A

PROJECT PHASE II
REPORT ON

“Agribot Insights Crop data-driven identification system”

Submittedto the

In infulfilment of the

requirements for the award of

the degree

BACHELOR OF TECHNOLOGY

COMPUTER SCIENCE AND ENGINEERING

2024 -2025
BY
Sneha Balappanavar 2164191242124
Mayur Adate 2164191242527

UNDER THEGUIDANCE OF

Dr. Sanjeevkumar Angadi

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

PCET-NMVPM’s
NUTAN COLLEGE OF ENGINEERING AND RESEARCH
TALEGAON, PUNE 410507
 PCET-NMVPM’s
NUTAN COLLEGE OF ENGINEERING & RESEARCH
TALEGAON, PUNE

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

CERTIFICATE

This is to certify that the Project Report entitled “Agribot insights data driven identification system ”, which
is being submitted by, Sneha Balappanavar , Mayur Adate as partial fulfillment for the Degree Bachelor of
Technology (Computer Science and Engineering) of DBATU, Lonere.
This is bonafide work carried under my supervision and guidance.

Place:Pune
Date:

Dr. Sanjeevkumar Angadi Mrs. Shital Mehta

Project Guide Project I/C

Dr. Sanjeevkumar Angadi Dr. Aparna Pande

Head of Department Principal

External Examiner [Name & Sign] SEAL

 ACKNOWLEDGEMENT

The satisfaction and euphoria that accompany the successful completion of any task would be

incomplete without the mention of people who made it possible. So, we acknowledge all those whose

guidance and encouragement served as a beacon light and crowned our efforts with success.

We have immense pleasure in expressing thanks to the principal Dr. Aparna Pande for

providing all the facilities for the successful completion of the project.

With due respect, we thank our Head of Department Dr. Sanjeevkumar Angadi, Computer

Science and Engineering, for his motivating support, keen interest which kept our spirits alive all

through.

We would like to express thanks to our guide Dr. Sanjeevkumar Angadi, Department of

Computer Science and Engineering who has guided us through out the completion of this project.

Finally, we would like to thank all the teaching and non-teaching staff and all our friends

who have rendered their support in the completion of this report.

Sneha Balappanavar [2164191242124]

Mayur Amar Adate [2164191242527]

i
 ABSTRACT

An agriculture chatbot is an AI-driven virtual assistant designed to support

farmers by providing timely, accurate, and personalized information to enhance
agricultural productivity and sustainability. By leveraging technologies such as Natural
Language Processing (NLP), machine learning, and deep learning, these chatbots can
interpret user queries and deliver relevant advice on various farming aspects. For instance,
the Agriculture Chatbot Voice Assistant (ACVA) employs Multi-Layer Perceptron (MLP)
neural networks and NLP techniques to analyze complex agricultural datasets. This
enables it to offer tailored guidance on crop management, pest control, and market trends.
Its voice recognition capabilities further enhance accessibility, allowing farmers to interact
seamlessly, even in remote or hands-free environments. To assist in various agricultural
tasks such as seeding, weeding, irrigation, harvesting, and crop monitoring. This project
presents the design and implementation of an Agribot capable of performing field
operations with minimal human intervention. Equipped with sensors, GPS, and machine
learning algorithms, the Agribot can navigate fields, detect crop health, and make real -
time decisions to optimize productivity. By integrating automation into agriculture, this
system aims to increase efficiency, reduce manual labor, and promote sustainable
farming practices.

Keyword – Machine Learning, Crop prediction ,Natural language processing ,

Artifical intelligence

ii
 LIST OF CONTENTS

CONTENTS PAGE NUMBER

Acknowledgment i
Abstract ii
List of content iii
List of table iv
List of Figures v
List of Abbreviation v

iii
Chapter Contents Page No
No
Introduction 1
1.1 Necessity 3
1
1.2 Problem Statement 6
1.3 Objective 7
1.4 Motivation 8
2 Literature Survey 9
System Analysis 11
3.1 Requirement Analysis 11
3.2 System Requirements (Hardware & Software) 13
3
3.3 Sequence diagram/DFD’s/ER Diagrams (as applicable) 16
3.4 Use case diagram 17
3.5 DFD Diagram 18
3.6 Class Diagram 21
Project Implementation 22
4.1 Introduction 22
4 4.2 Tools and technologies used 22
4.3 System Architecture 23
4.4 Methodolgy and algorithms used 26
Software Testing 33
5 5.1 Types of Testing used 33
5.2 Working of testing 35
5.3 Test case result 38
6 Results and Discussion 39

6.1 Introduction 39
6.2 Evaluation matrix 39
6.3 CNN model performance 40
6.4 System output example 43
Conclusion and Future Scope 48
References 50
 LIST OF FIGURES

Figure No. Name of Figure Page No.

1.1 Crop prediction 2
3.1 Sequence diagram 16
3.2 Use case Diagram 17
3.3 DFD 0 18
3.4 DFD 1 18
3.5 DFD 2 20
3.6 Class Diagram 21
4.1 System architecture diagram 25
4.2 SVM working 29
4.3 CNN working 32
5.1 Stages of testing 36

6.1 Parallel CNN accuracy 40

6.2 Confusion matrix of parallel CNN 41
6.3 Svm algorithm 42
6.4 Login page of chatbot and crop prediction 43
6.5 Interface of chatbot 44
6.6 Interface of crop prediction 45
6.7 Interface of Svm prediction and training 46

iv
 LIST OF TABLES

Table No Name of Table Page No

2.1 Literature survey table 9

3.1 Hardware requirement 13

3.2 Software requirement 14

4.1 Integration in agribot 29

4.2 Use case of crop predication and chatbot 31

5.1 Type of testing 33

5.2 Test case 36

6.1 Evaluation matrix 37

LISTOF ABBREVIATIONS

Sr. No Abbreviations Description

1 CNN Convolutional Neural Network

2 ANN Artificial Neural Networks

3 SVM Support Vector Machine

4 AI/ML Artificialintelligence /Machine

learning

v
 Agribot insights data-driven crop identification system

CHAPTER 1 . INTRODUCTION
Agriculture has always been a cornerstone of human civilization,
providing the food and raw materials necessary to sustain life and drive economies.
In many countries, especially developing nations, the agricultural sector employs a
significant portion of the population and contributes substantially to national GDP.
However, farmers today face a multitude of challenges that threaten productivity
and profitability. These challenges range from unpredictable weather patterns, pest
infestations, and lack of access to modern farming techniques, to inadequate market
intelligence and limited communication with experts. In this context, technology
emerges as a powerful enabler to overcome these obstacles. Over the past few
decades, advancements in Artificial Intelligence (AI) and Machine Learning (ML)
have transformed industries, offering innovative solutions to long-standing
problems. When applied to agriculture, these technologies can revolutionize
farming practices, making them more efficient,
The advent of AI-powered chat applications tailored specifically for
farmers is a groundbreaking step in this direction. These applications leverage
cutting-edge technology to bridge the gap between farmers and the information they
need to make informed decisions. Acting as virtual agricultural assistants, they
provide real-time support on a wide array of topics, including weather forecasts,
pest and disease management, crop selection, irrigation planning, soil health
analysis, and market price updates. One of the primary benefits of such an
application is its ability to offer personalized solutions. By analyzing data from a
farmer's location, crop type, and farming practices, the AI system can deliver
insights tailored to their unique needs. For example, a farmer growing rice in a
flood-prone region might receive specific advice on water management and disease
prevention, while a farmer cultivating cotton in a dry area could get
recommendations on irrigation and pest control. Another critical aspect of the
farmer's chat application is inclusivity. In many rural regions, language barriers and
digital literacy remain significant challenges. By incorporating support for multiple
regional languages and user-friendly interfaces, these applications ensure
accessibility for a diverse range of farmers.

Department of CSE, Nutan College of Engineering and Research, Talegoan, pune 1

 Agribot insights data-driven crop identification system

Even those with minimal technological experience can engage with the
platform, asking questions in their native language and receiving clear, actionable
response. Beyond individual support, the broader implications of AI in agriculture
are profound.
By aggregating data from multiple users, the system can contribute to large-scale
agricultural planning, helping governments and organizations identify patterns and
allocate resources effectively. For instance, if many farmers report similar pest
problems, the system can flag the issue for regional pest control initiatives. The farmer's
chat application is not just a tool; it represents a paradigm shift in the way agricultural
knowledge is disseminated and applied. Crop yields can be predicted using machine
learning methods like Random Forest, Support Vector Machines (SVM), and Artificial
Neural Networks (ANN). By examining past data, these algorithms find trends that can
be utilized to predict agricultural yields in the future

Fig 1.1 Crop prediction

The above fig (1.1) Crop prediction plays a vital role in modern agriculture by providing
accurate forecasts of crop yields based on environmental conditions, soil properties, and
agricultural practices. With the advancement of machine learning (ML) and artificial
intelligence (AI), predictive models can analyze vast datasets including temperature, rainfall.

Department of CSE, Nutan College of Engineering and Research, Talegoan, pune 2

 Agribot insights data-driven crop identification system

1.1 Necessity
In today's The agricultural sector is fundamental to the global economy, food
security, and rural development, but it faces numerous challenges that hinder
productivity and sustainability. Despite being an essential industry, farmers around
the world, particularly in developing regions, continue to grapple with outdated
practices, lack of access to resources, and inconsistent support. The necessity for
an AI and ML- powered farmer’s chat application arises from the urgent need to
address these challenges and revolutionize the way farmers operate. The modern
agricultural landscape demands technological solutions that keep pace with both
growing population needs and environmental sustainability goals.

Below, we explore in detail why such an application is critical, Agriculture

has long been at the heart of human development, supporting livelihoods for
billions and contributing significantly to global GDP. However, as the world faces
challenges such as climate change, dwindling natural resources, and population
growth, the pressure on farmers to increase productivity without compromising
environmental health has never been greater. Farmers must navigate this complex
environment with minimal tools and resources, making the integration of
technology essential for meeting future demands. An AI and ML- powered farmer’s
chat application can be the game-changer that bridges the current gap in agricultural
practices and modern needs. Traditional farming methods, though often tried and
true, cannot match the precision, efficiency, and adaptability that modern
technology can offer. While some farmers have adopted better equipment and
practices, many still face hurdles related to the cost, availability, and training
required to integrate such innovations. For the millions of smallholders, particularly
in rural areas of developing countries, the challenges are amplified by limited
access to reliable information, outdated methods, and a lack of support systems that
are scalable and inclusive. An AI and ML- powered chat application can provide
real-time information and personalized advice, addressing these gaps and leveling
the playing field for farmers regardless of their geographic location or
socioeconomic status. One of the most significant challenges in farming today is
the ability to access accurate, timely, and relevant information. Weather forecasts,
pest management techniques, crop rotation best practices, and market trends are
just a few of the critical areas where up-to-date knowledge can make a substantial
difference.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 3

 Agribot insights data-driven crop identification system

Additionally, resource management has become a pressing issue, particularly

concerning water and chemical usage. Excessive use of water and pesticides not
only strains natural resources but can also result in long-term environmental
damage. Sustainable farming practices, such as precision agriculture, are crucial to
addressing these issues, and AI and ML can play a vital role in achieving them.
24/7 Customer Support: AI chatbot can provide instant customer support 24/7,
which means that farmers and customers can get help at any time, even outside of
business hours.
Cost-effective: AI chatbot can handle multiple customer queries at the same time,
which reduces the need for human customer service agents. This can lead to
significant cost savings for businesses in the agriculture industry.
Personalized Experience: AI chatbot can use NLP and ML to understand customer
queries and provide personalized responses. This improves the customer experience
and helps to build brand loyalty.
Improved Efficiency: AI chatbot can handle repetitive and simple queries, which
frees up human customer service agents to handle more complex queries. This
improves the efficiency of the customer service team.
Data Analysis: AI chatbot can collect data on customer queries and interactions,
which can be used to improve the customer experience and identify areas of
improvement.
Farmers often struggle to access accurate and up-to-date information. While
agricultural advisors and extension services are valuable, their reach is limited, and
their availability is inconsistent, especially in remote areas. Farmers may face
significant delays when waiting for expert consultations, leading to lost
opportunities and preventable crop failures. An AI- powered chat application can
provide immediate access to expert-level guidance, enabling farmers to make
timely decisions regarding pest control, irrigation, fertilization, and crop rotation.
This rapid response capability is essential for minimizing the risks associated with
modern farming and enhancing overall productivity.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 4

 Agribot insights data-driven crop identification system

 Pest and Disease Management

Pests and diseases can devastate crops and significantly impact a farmer’s
livelihood. Traditional methods of pest control and disease management are often
reactive, leading to severe yield losses and financial strain. An AI-driven chat
application can help farmers detect and diagnose diseases and pest infestations early
by using image recognition and data analysis. The application can then provide
farmers with preventive measures and treatment plans, thereby reducing damage
and promoting healthier crops. This proactive approach is crucial for minimizing
crop losses and ensuring sustainable agricultural practices.

 Unpredictable Weather and Climate Change

Climate change has brought unpredictable and extreme weather patterns, making it
increasingly difficult for farmers to plan their farming activities. Accurate weather
forecasts and timely updates are vital for managing activities such as sowing,
irrigation, and harvesting.

 Resource Optimization and Sustainability

The overuse of water, fertilizers, and pesticides has led to environmental
degradation, water scarcity, and declining soil health. Sustainable farming practices
that conserve resources and maintain ecological balance are essential for long-term
agricultural success. An AI and ML- driven chat application can help farmers
optimize the use of water and fertilizers based on soil conditions and crop needs,
thus minimizing waste and promoting better resource management. It can also
recommend eco-friendly practices, such as crop rotation and organic farming,
contributing to environmental sustainability and enhancing soil health over time.to
lost opportunities and preventable crop failures. An AI-powered chat application
can provide immediate access to expert- level guidance, enabling farmers to make
timely decisions regarding pest control, irrigation, fertilization, and crop rotation.
This rapid response capability is essential for minimizing the risks associated with
modern farming and enhancing overall productivity.
 Market Awareness and Fair Pricing
Farmers often face challenges in accessing current market prices and trends, leading
to a reliance on intermediaries who may exploit them by offering lower.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 5

 Agribot insights data-driven crop identification system

1.2 Problem statement

In Agriculture is the backbone of human civilization, providing food
security, employment, and economic stability for billions of people worldwide.
Despite its critical importance, the sector faces severe challenges that limit
productivity, sustainability, and profitability, particularly in developing regions.
Farmers, especially small holders and those in rural areas, often grapple with the
dual burden of outdated practices and limited access to modern resources and
technologies.

This creates a pressing need for solutions that bridge the gap between
traditional methods and the demands of modern agriculture. One of the primary
issues in agriculture is the lack of access to reliable, timely, and expert
knowledge. Coupled with this is the inability to access real-time information,
such as accurate weather forecasts, pest outbreak alerts, or market price trends.
Farming is highly time- sensitive, and delayed or incorrect information can result
in significant crop losses and reduced incomes . Market access remains another
critical challenge for farmers. Many smallholders are forced to rely on
middlemen to sell their produce, which often leads to exploitation. These
intermediaries purchase crops at rates significantly below market value,
depriving farmers of fair profits. The lack of direct access to markets, coupled
with limited bargaining power, exacerbates their economic vulnerability, leaving
them unable to invest in better equipment, seeds, or sustainable farming
methods. In addition to market inefficiencies, the management of natural
resources like water, fertilizers, and pesticides is often inefficient. Farmers
frequently overuse these inputs, leading to increased costs, environmental
degradation, and long-term harm to soil health.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 6

 Agribot insights data-driven crop identification system

1.3 Objectives
The main goal of the proposed AI and ML-powered farmer’s chat
application is to bring significant advancements to the agricultural sector,
empowering farmers with tools and resources that can enhance their
productivity, profitability, and sustainability. This platform aims to address the
challenges faced by farmers by integrating modern technology with traditional
farming methods to bridge the knowledge and resource gap that currently exists.
 A fundamental objective is to enhance access to expert knowledge for farmers,
especially those in remote and underserved areas.
 The application will offer direct, real-time guidance on critical aspects of
farming, such as pest management, soil health, crop selection, and disease
control
 Providing real-time, actionable information is also at the core of this platform’s
objectives. The AI and ML algorithms will enable farmers to receive up-to-
date weather forecasts, pest alerts, and market price trends.
 The application will facilitate direct market access, giving farmers the ability
to connect with buyers and markets without the need for middle men.
 A crucial aspect of the application’s design is the promotion of efficient
resource management.
 The application’s reach will extend beyond technology access to include
bridging the digital and knowledge divide that exists within farming
communities.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 7

 Agribot insights data-driven crop identification system

1.4 Motivation
 The motivation behind developing an The motivation behind developing the
AI and ML- powered farmer’s chat application stems from recognizing the
pressing challenges that farmers face and the transformative potential that
technology holds in addressing these challenges. Agriculture, while being a
crucial sector for global food security and economic stability, is often
overlooked when it comes to advancements in technology. Farmers,
especially those in rural and underserved regions, continue to struggle with
outdated farming practices, limited access to information, and economic
constraints that hinder their productivity and quality of life. This realization
has spurred the drive to create a platform that can empower farmers with the
resources they need to succeed.
 One of the main motivations for this project is the desire to bridge the gap
between traditional farming
 methods and modern agricultural practices. While technology has
revolutionized many industries, the agricultural sector has been slower to
adopt these changes. The integration of AI and machine learning into farming
can provide insights that were once out of reach, enabling farmers to make
better, data-driven decisions. The potential benefits of these technologies
include improved crop yields, reduced costs, and more efficient resource
management, all of which are vital for the long- term success of farming
communities.

 Farmers often face an information gap that limits their ability to make
informed decisions about everything from soil management and crop
selection to pest control and market pricing. This lack of access to reliable
knowledge can result in poor farming practices, reduced productivity, and
economic hardship. By providing a platform that offers expert advice and
real-time data, the application aims to empower farmers to optimize their
operations and improve their outcomes. This motivation is driven by a belief
that every farmer, regardless of location or literacy level, should have the
opportunity to harness the power of modern technology to enhance their
productivity and sustainability.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 8

 Agribot insights data-driven crop identification system

CHAPTER 2. LITERATURE SURVEY

Table 2.1 Literature survey

Sr Paper Autor Method/ Advantage Disadvantage

No
Name, Name Techniqe
Year

1 AI-
Powered D.Katiyaret AI Enhances High
al. chatbots farmer resource
Agriculture
accessto requireme
Chatbots advice nts,
and limited
for
market connecti
Farmers, insights viy
2024 [1]

2 Agricultural
V.Himabindu AI- Versatile Limited
Helper et al. based crop specific
Chatbot, chatbot managem query
nt and handling
2023 [2] pest
control

3 Farm
K.Mandalet NLP Improves Language
Friendly al. chatbot farmer- barriers,nee
Chat Bot, expert d for
updates
2023 [3]
4 The
Overview General,
digitization Abbas Literature
of AI lacks
of iet al. review advancem technical
e nt s in specifics
agricultural agriculture
industry,
2022 [4]

5 Deep Requires
Adkeetal. Precise high-quality
learning Deep
crop imagery
for learning, analysis,
efficiency
cotton image

boll processing

analysis,
2022 [5]

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 9

 Agribot insights data-driven crop identification system

6
Knowledge Knowl Enhances High
Agarwa
chatbot computationa
graph e dge
l et al. understandi l demand
pretraining graph, -ng
, 2021 [6] languag
e
models

7
Accuracy Ahlers,D. Reliable Limited
Geospati
location direct
ofgeo
al data data for application
names farming
assessm
data, ent
2013 [7]

8
Improved Specificto
Cranberr Akivae Image
crop berry
segmentatin, t al. segmen monitoring crops
2020[8] tation

9
AI in Alam,A.el AI Shows AI Not field-
education learning educationa specific,rural
,2021 [9] analytics l benefits barriers

10
Agribot Katherine Python Predict May be ,
(2018) el ,AI,ML,nl crope difficultfor
[10] asily farmer
sto use

11
Farmchat(2019) Antonyparr Diaglow Predicts Specificto
[11] impleme crop berrycrop
mtation according
to
enviromen-
t
12
Krishibot Krishnaswami BERT, Enhance Related
[12] farmers to
Google,
Dia logflow experience particular
, Regionbased
mangodb onlanguage

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 10

 Agribot insights data-driven crop identification system

CHAPTER 3. SYSTEM ANALYSIS

3.1 Requirement Analysis

The development of an AI and ML-powered farmer's chat application involves a
synergy of advanced technologies, each playing a vital role in creating a
comprehensive, efficient, and user-friendly platform. These technologies
collectively ensure the system can process diverse types of data, interact naturally
with users, and provide actionable insights that enhance farming practices. Below
is an in-depth exploration of the key technologies utilized, showcasing their
implementation, advantages, and relevance to the agricultural domain.

Functional Requirements
These define the specific behaviors and functions the agricultural chatbot must perform:
 User Interaction
The chatbot should allow users, including farmers and agronomists, to interact via
both text and voice inputs. Support for regional and local languages is essential to
ensure accessibility.

 Query Handling
The system must be capable of answering agricultural queries related to crops, soil
types, fertilizers, pesticides, and weather conditions. It should also provide real-
time weather forecasts and alerts.

 Crop Diagnosis
Users should be able to upload images of crops showing symptoms of disease or
pest attacks. The chatbot should analyze the image and suggest potential treatments
or preventive measures.

 Notifications and Reminders

It should send timely reminders for critical farming tasks such as irrigation,
fertilization, sowing, or harvesting based on user data and seasonal trends.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 11

 Agribot insights data-driven crop identification system

Non-functional Requirements
These define how the system should perform its functions:

 Performance
The chatbot should deliver responses quickly, ideally within 2 to3 seconds for most
queries.

 Scalability
It must be capable of supporting a large number of concurrent users, particularly
during peak .

 Reliability and Availability

The system should be highly reliable with a minimum uptime of 99.9% and resilient
to issues like poor network conditions in rural areas.

 Security and Privacy

User data, such as location and farm details, must be protected using encryption. The
system must comply with relevant data protection regulations.

 Usability
The chatbot interface should be simple and user-friendly, catering to users with
minimal technical knowledge. It should also be accessible in low-bandwidth areas.

 Maintainability
The system should be easy to update and maintain, allowing for quick bug fixes
and enhancements without extended downtime.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 12

 Agribot insights data-driven crop identification system

3.2 System Requirement

Table 3.1 Hardware requirement

Component Specification
Processor Multicore CPU or GPU for handling
computationally intensive tasks such as
deep learning and image processing

RAM At least 8GB, preferably 16GB or efficient

handling of high-resolution image and
videos

Storage Sufficient disk space (minimum 500GB)

for storing datasets, processed output, and
intermediate results

Camera system High-quality cameras capturing

images or videos in various
environmental conditions for real-
world testing

Sensors Monitor the water content in the soil to

optimize irrigation. Track environmental
conditions affecting crop growth. Assess
soil acidity or alkalinity, crucial for
nutrient management. Measure soil
nutrient levels (Nitrogen, Phosphorus,
Potassium).

Actuators and Mobility Components Provide movement capabilities for

wheels or tracks. Enable precise
control of tools like seed dispensers or
robotic arms. Control the direction
and speed of motors.

Communication modules Facilitate short-range wireless communication

with smartphones or tablets. Enable internet
connectivity for remote monitoring and data
logging.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 13

 Agribot insights data-driven crop identification system

Table 3.2 Software requirement

Component Specification

Operating System Windows10(64 Bit)

IDE Spyder

Programming Language Python version3.7, 3.8

Libraries Tensor Flow ,Open CV ,Keras,NumPy

Database DBSqlite3

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 14

 Agribot insights data-driven crop identification system

 Once the system processes the data and applies the relevant models, it generates a
response. This response is context-aware, addressing the farmer’s concerns while offering
practical recommendations. For instance, if a pest problem is detected, the application
may suggest an appropriate pesticide or a natural remedy, along with usage instructions.
The response may also consider the emotional tone of the farmer’s query, ensuring
that it is empathetic and reassuring when needed.
 The application’s design includes continuous learning mechanisms. Each interaction
contributes to the system’s improvement through feedback loops, where farmers can
rate the relevance.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 15

 Agribot insights data-driven crop identification system

3.3 Sequence diagram

Fig 3.1 sequence diagram of crop predication

The provided (fig3.1) class diagram illustrates the structure of a chat-based

communication system, encompassing three main classes: Chat based Communication,
User, and ChatBot. The Chat- Based Communication class serves as a central container,
managing lists of users and chat bots within the system. Users, represented by the User
class, possess unique identifiers (user Id) and display names (username). Users can send
messages through the send Message method.
On the other hand ,the ChatBot class embodies individual chat bots, characterized
by unique identifiers (bot Id), display names (bot Name), and an activity status (is
Enabled). Chat bots also employ the send Message method. The relationships are
expressed through aggregation, signifying that the Chat Based Communication class
contains instances of users and chat bots. While this diagram provides a foundational
representation, real-world applications may necessitate further refinement, incorporating
elements like message handling, user authentication, and additional features to align with
specific system requirements.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 16

 Agribot insights data-driven crop identification system

3.4 Use case Diagram

Fig 3.2 case diagram for crop predication

Actors and Their role on use case diagram

o User: Initates the process by providing the image capturing system

o Image processing: Responsible for capturing image frames from the input image.
o Analysis of plant detection : By using algorithm, and dataset it predicts the crop.
o Storage Folder: Stores the extracted frames temporarily.
o Result : Creates the output image file by continuing the using algorithm

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 17

 Agribot insights data-driven crop identification system

3.5 DFD Diagram

Fig 3.3 DFD 0

Fig 3.4 DFD 1

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 18

 Agribot insights data-driven crop identification system

The above (fig.3.3) and (fig.3.4) Data Flow Diagram (DFD 0 & 1 ) for the
farmer's chatbot revolves around several core elements: user, dataset, response,
questions, answers, category, role, and identity, each playing a crucial part in the
system's functionality. The user, in this case, is the farmer, who interacts with the
chatbot by asking questions related to agricultural activities. These interactions are
initiated in various forms, such as text-based queries, voice inputs, or even images
(e.g., for pest identification).
The chatbot uses this input to identify the farmer's intent and determine the
category of the question, such as crop health, market prices, weather updates, or
general agricultural advice. The chatbot processes the farmer's questions by
categorizing them into relevant topics. Once categorized, it fetches the required
information from the respective datasets. The chatbot also considers the user's
identity to tailor its responses. Identity details, such as the user’s location and
preferred language, ensure that the advice is region-specific and easy to understand.
For instance, a farmer in Maharashtra asking about pest control for sugarcane will
receive tailored guidance applicable to that crop and region.
Once the data is retrieved, the system formulates a response. These
responses are crafted using advanced natural language processing (NLP) and
machine learning algorithms. They are then delivered in the user’s preferred
format—be it text, voice, or visuals. For example, when asked about fertilizer
recommendations, the chatbot might respond with detailed textual advice or images
showing application techniques. If the user queries about market prices, the chatbot
fetches real- time pricing data from its market dataset and S delivers it concisely.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 19

 Agribot insights data-driven crop identification system

A (fig.3.5) Data Flow Diagram (DFD Level 2) shows more detailed internal processes of
a system, focusing on how data flows within sub-processes of a Level 1 DFD. Here's a Level 2
DFD for an Agriculture Chatbot System, broken into key modules like User Interaction, Crop
Advisory, Market Info, and Admin Management. The system allows farmers to ask questions or
request services via chatbot (text or voice). It connects to external APIs (weather, market), a database
(crop info, farmer profiles), and can involve admin/moderator review.

Data Flow Overview:

Farmer →2.1 Handle User Input → Text or Voice Query
2.1 → 2.2 NLP & Intent Detection→ Determine intent (ask about crop, weather, etc.)
2.2→2.3 Crop Advisory (uses D2 Crop KB &D1Farmer Profile)
2.2→2.4 Market / Weather Info(calls External APIs& uses D3 for cache)
2.2 → 2.5 Admin Support (If query not resolved)
2.6 logs feedback or updates to D1, D4

Fig 3.5 DFD 2

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 20

 Agribot insights data-driven crop identification system

3.6 Class Diagram

Fig 3.6 Class Diagram of Crop predication Working

The (fig 3.6 ) Farmers Bot class inherits from the Chatbot class and extends its functionality
to cater specifically to farmers. For instance, while the Chatbot class has a generic fetch Data()
method, Farmers Bot uses its specialized datasets (like market trends and crop-specific data)
to provide more precise and relevant information.
This structure demonstrates the modular and extensible design of the system, allowing the
Farmers Bot to leverage the general capabilities of the Chatbot while providing domain-
specific functionalities for farmers

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 21

 Agribot insights data-driven crop identification system

CHAPTER 4 . PROJECT IMPLEMENTION

4.1 Introduction
Crop yields can be predicted using machine learning methods like Random Forest,
Support Vector Machines (SVM), and Artificial Neural Networks (ANN). By examining
past data, these algorithms find trends that can be utilized to predict agricultural yields in
the future. There are various advantages of using machine learning to anticipate crops.
whether it comes to crop management, including whether to plant, fertilize, or irrigate
crops, farmers may make better decisions with greater knowledge.

4.2 Tools and Technology used

1. Tensor Flow/Keras – to support BERT and deep learning components
2. Pandas, NumPy – for data handling and manipulation
3. Programming Language: Python
4. Scikit-learn–for implementing traditional ML models like SVM, XGBoost

5. Matplotlib, Seaborn – for result visualization

6. Database: SQLite – used for storing user details and system logs

7. Hugging Face Transformers – for pre-trained BERT model fine-tuning

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 22

 Agribot insights data-driven crop identification system

4.3 System Architecture

 SVM: Classifies news as fake or real using extracted vectorized features.
 NLP: Natural language processing
 Visualization: Accuracy, F1-score, confusion matrix, and
misclassification reports are generated.
 User Interface (GUI): Built using Tkinter for login, registration, and
detection interface.

1. Hardware Components
 Microcontroller Unit (MCU): Acts as the central processing unit, often
utilizing platforms like Arduino, Raspberry Pi, or AVR microcontrollers.

 Sensors: Includes temperature and humidity sensors (e.g., DHT 11), soil
moisture sensors, ultrasonic sensors for obstacle detection, and light sensors to
monitor environmental conditions.

 Actuators: Comprises motors for movement, solenoid valves for irrigation

control, and mechanisms for seed dispensing and pesticide spraying.

 Power Supply: Typically powered by rechargeable batteries, with some models

incorporating solar panels for sustainable energy.

 Communication Modules: Utilizes Bluetooth (e.g., HC-05), Wi-Fi, or GSM

modules to facilitate remote control and data transmission.

2. Software components

 Control Algorithms: Embedded software governs the robot's

operations, including navigation, task execution, and sensor data processing.

 User Interface: Often features a mobile or web-based application enabling

users to send commands, monitor status, and configure settings.

 Data Processing: Incorporates algorithms for real-time decision-making based

on sensor inputs, such as adjusting irrigation schedules or detecting obstacles.

 Cloud Integration: Some systems connect to cloud platforms for data

storage, analytics, and remote monitoring, enhancing scalability and
accessibility.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 23

 Agribot insights data-driven crop identification system

3. Functional Modules

 Navigation System: Employs GPS and sensor data to autonomously

navigate fields, ensuring coverage and avoiding obstacles.

 Task Execution Units: Dedicated modules handle specific tasks like

plowing, seeding, watering, and pesticide application, often configurable
based on crop requirements.

 Monitoring and Feedback: Continuously collects environmental data to

assess conditions and adjust operations accordingly, promoting precision
agriculture

Fig 4.1 System architecture Diagram of crop prediction

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 24

 Agribot insights data-driven crop identification system

The above (fig.4.1) Chatbots are AI-driven applications designed to simulate human-
like conversations, providing users with instant responses and assistance. The core
functionality of a chatbot involves processing user inputs, understanding intent,
retrieving relevant information, and delivering appropriate responses. This process
leverages Natural Language Processing (NLP) and machine learning algorithms to
interpret and respond to user queries effectively.

The (fig 4.1) diagram above outlines the typical work flow of a chatbot system

 Start: The chatbot session is initiated.

 User Input: The user sends a message or query to the chatbot.
 Preprocessing: The chatbot pre processes the input, which may
include tokenization, removing stop words, and other NLP tasks.
 Intent Recognition: The system analyzes the processed input to
determine the user's intent.
 Entity Extraction: Relevant entities or keywords are extracted from the
input to provide context.
 Response Generation:
 If the intent and entities match known patterns, the

chatbot retrieves the appropriate response from its

knowledge base.
 If the input is ambiguous or doesn't match known patterns, the

chatbot may ask the user for clarification.

 Response Delivery: The chatbot sends the generated response back to the user.
 End: The session continues or ends based on user interaction.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 25

 Agribot insights data-driven crop identification system

4.4 Methodology/Algorithms details

 Support Vector Machine (SVM): Used for high-dimensional
textual data. Trained using TF-IDF and n-gram features.

 Image Processing &Crop Analysis -Used for plant crop prediction

, weed recognition, fruit counting, etc.

 Convolutional Neural Networks (CNNs) – e.g., Res Net, Efficient Net

for image classification.

 YOLO (You Only Look Once) – For object detection like fruits,
weeds, pests.

 Semantic Segmentation – e.g., U-Net

The development of Agribot follows a structured methodology combining

machine learning, image processing, and database integration. Real- time image
data is captured using cameras and processed using OpenCV for noise reduction,
segmentation, and feature extraction.
These processed images are then fed into deep learning models built with
TensorFlow and Keras to detect crop diseases, pests, and anomalies. The robot's
decision-making logic is implemented in Python, enabling it to perform actions
such as spraying or sending alerts. All sensor readings, image analysis results, and
system logs are stored locally using DBSqlite3 for efficient data management and
later analysis.

1) SVM

1. Terrain Classification in Agribots

Agribots often traverse varied terrains (e.g., vegetation, sand, gravel). By embedding
SVMs that process accelerometer or vibration signals, they can predict terrain type to
adjust movement:

 Input data: Vibration/acceleration readings captured by sensors.

 Feature extraction: Identify patterns—e.g., mean, variance ,frequency components

unique to each terrain.
 SVM model:

o Trained with labeled samples from different terrains.

o Explores linear, polynomial, or RBF kernels to manage non-linear separable data.
o Uses 10-fold cross-validation to optimize performance.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 26

 Agribot insights data-driven crop identification system

 Outcome: The SVM classifies each terrain in real time, enabling the robot to adapt
its locomotion.
 Example: A study embedding SVM with accelerometer data could distinguish
vegetation, sand, and gravel effectively

2. Crop prediction via Image Processing

Agribots also analyze plant leaf images to detect diseases, employing SVMs as
classifiers:

 Image features: Techniques like Histogram of Oriented Gradients (HOG) are used
to extract shape and texture information. 
 SVM classification: The model (often with RBF kernel) separates ' classes based
on the extracted features.
 Example performance: A hybrid HOG + SVM model detected tomato late-blight
with high precision (evaluated against Decision Trees and KNN), showing SVM's
effectiveness in crop prediction

3. How SVM Works – The Mechanics

 SVM transforms input features into higher-dimensional space using a kernel

function (e.g., linear, polynomial, RBF).

 Searches for the maximum-margin hyperplane that best separates the classes
by solving:
 Once trained, new data is classified by determining on which side of the
boundary it lies. 

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 27

 Agribot insights data-driven crop identification system

4. Integration into Agribot Systems

Table 4.1 Integration in agribot

Step Description

Accelerometers for terrain; cameras for plant

Sensors
imaging.

Feature Signal stats or HOG descriptors feed into the

Extraction SVM.

Model SVM trained offline using labeled

Training datasets; kernel and parameters tuned via
cross-
validation.
Real-time classification embedded in the robot’s
Deployment
control loop.

This enables agribots to navigate optimally, apply fungicides precisely, and protect
plants early, all autonomously.

Fig 4.2 SVM working

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 28

 Agribot insights data-driven crop identification system

The above (fig 4.2) Support Vector Machines (SVM), including Support Vector
Regression (SVR), provide a robust framework for predicting crop yields and
classifying crop types and conditions. These methods are especially effective when
dealing with small- to medium-sized datasets and high-dimensional feature spaces, such
as those arising from soil variables, climatic records, and remote sensing inputs One
notable study introduced a multi-attribute weighted tree-based SVM (MAWT-SVM),
which integrates z-score normalization, Principal Component Analysis (PCA), and
genetic algorithm (GA)-driven feature selection, achieving top-tier performance in crop
yield prediction SVMs have also been successfully combined with other models. A
hybrid framework using XGBoost, SVM, was shown to effectively predict crop yields
based on temperature, rainfall, and soil characteristics, delivering strong results across
multiple statistical metrics
In remote sensing contexts, SVR using satellite-derived vegetation indices—often with
RBF kernels—has delivered accurate yield predictions. For example, field-scale rice
yield forecasting via Sentinel-2 imagery found SVM outperforming Random Forest and
ANN, with RMSPE values around 5–11% depending on the season

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 29

 Agribot insights data-driven crop identification system

2) CNN

 Spatial feature extraction

 CNNs automatically learn spatial patterns (e.g., canopy structure, soil texture)
from UAV, satellite, or ground images using stacked convolution + pooling
layers

 Temporal & hybrid modelling

 1D-CNNs process environmental variables (weather/soil) as sequences. They
are often combined with RNNs/LSTMs to capture temporal trends. 
 E.g., the CNN–RNN model predicted corn and soybean yields in the U.S.
with RMSE ~8–9% of average yields, outperforming traditional methods

 Performance Gains

 In Germany, a 1D-CNN model on winter wheat reduced RMSE by7–14%,

MAE by 3–15%, and boosted correlation by 4–50% versus baselines

 UAV-based CNNs using RGB imagery achieved early-season yield

forecasts with MAE ~484 kg/ha (MAPE ~8.8%)

 Multimodal & attention-based designs

 Advanced models like BO-CNN- BiL STM include attention mechanisms and
Bayesian tuning for better integration of remote sensing and climate data CNN-
LSTM hybrids with attention have achieved R² ≈ 0.967 for Indian wheat and rice
(RMSE 0.017, MAE 0.09)

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 30

 Agribot insights data-driven crop identification system

Fig 4.3 CNN working

The above (fig 4.3) Convolutional Neural Networks (CNNs), often in combination with
time-series models or attention mechanisms, have emerged as powerful tools for crop
yield estimation and classification tasks in precision agriculture. These models leverage
spatial features from remote sensing imagery (e.g., UAV-/satellite-derived NDVI,
RGB, multispectral data) and temporal environmental variables (e.g., weather records,
soil, and phenological measurements) to accurately forecast yields across diverse crops
and regions.
CNN-based methods excel at feature extraction—automatically learning spatial and temporal
dependencies. CNN–RNN hybrids further model growth dynamics over time (e.g. genotype
improvements, seasonal variation). Integrating attention mechanisms refines focus on
influential features. These models consistently outperform classical machine learning
alternatives in yield prediction accuracy.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 31

 Agribot insights data-driven crop identification system

Table no 4.2 Use case of crop prediction and chatbot

Use Case Model Type Strengths

Crop CNN or CNN–RNN Learns spatial and temporal visual patterns from
Predictio hybrids imagery+ time-series data
n

Chatbot Transformers Models long-range language dependencies and generates

s (text) (GPT- type) coherent responses

 Chatbots today primarily use transformer-based LLMs (like GPT), not CNNs

 Self-attention layers to model complex language dependencies.

 Transformer architectures, fine-tuned on conversational or domain-specific
data.
 Pre-training + fine-tuning, enabling general language understanding and
generation.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 32

 Agribot insights data-driven crop identification system

CHAPTER 5. SOFTWARE TESTING

5.1 Type of software testing

 Unit Testing
It is the testing of individual software units of the application .it is done after the
complexion of an individual unit before integration. Unit testing involves the
design of test cases that validate that the internal program logic is functioning
properly, and that program inputs produce valid outputs.

 Regression Testing

Regression testing is a software testing practice that ensures an application still

functions as expected after any code changes, updates, or improvements. Regression
testing is responsible for the overall stability and functionality of the existing
features.

 Smoke Testing
Smoke Testing comes into the picture at the time of receiving build software
from the development team. The purpose of smoke testing is to determine
whether the build software is testable or not. It is done at the time of ―building
software.‖ This process is also known as ―Day 0‖.

 Testing Strategy
Software testing methods are traditionally divided into white- and black-box
testing. These two approaches are used to describe the point of view that a test
engineer takes when designing test cases.

 White-box testing
In white-box testing an internal perspective of the system, as well as
programming skills, are used to design test cases.

 Black-box testing
Black-box testing treats the software as a quot black boxquot; examining
functionality without any knowledge of internal implementation. The testers are
only aware of what the software is supposed to do, not how it does it.

 Grey-box testing
Grey-box testing involves having knowledge of internal data structures and
algorithms for purposes of designing tests, while executing those tests .

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 33

 Agribot insights data-driven crop identification system

Table no 5.1 Type of testing

Testing Type Crop Prediction Chatbot

Data transformations, feature

Unit Testing NLP components, intent parsers
engineering

Integration Testing Full data-to-prediction pipelines API, DB, service integrations

Functional/ Model Cross-validation, hold-out, metric Intent-response correctness,

Testing tracking conversation flows

E2E Testing Deployment scenario simulation Full conversation simulations

Regression Testing Use-case validation after model Re-run test suites post update
updates

Performance/ Load
Monitor predictions speed, scale Load simulation under concurrent use
Testing

Security Testing Data privacy, input validation Penetration tests, DAST

Ad-hoc / Fuzz Testing Random input generation, error Random/edge-case conversation testing
resilience

Model interpretability, UX testing, user trials,

Usability/ UAT / A/B
field feedback A/B conversational flows

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 34

 Agribot insights data-driven crop identification system

5.2 Working of Testing

1. Data Splits & Validation Strategy

 Training / Validation / Test Splits

Initial data division (e.g., 60–80%training, 20–40% testing) helps in evaluating
performance on unseen data during development .

 Cross-Validation (e.g., k-fold, leave-one-year-out)

Enhances reliability of performance estimates by evaluating models across multiple
data subsets—especially crucial for small agricultural datasets and time-varying
patterns
2. Performance Metrics
Crop prediction is typically a regression problem, evaluated using:

 MAE (Mean Absolute Error)

 RMSE (Root Mean Squared Error)
 R² (Coefficient of Determination)
 MAPE (Mean Absolute Percentage Error)
These metrics quantify inaccuracies between predicted and actual yield values

3. Spatial& Temporal Testing

 Geographical validation
Models are tested on regions absent from training data to assess generalizability
across different agro-climates

 Temporal validation
Techniques like leave-one-year-out or rolling-window cross-validation prevent
data leakage and ensure performance over time

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 35

 Agribot insights data-driven crop identification system

4. Uncertainty & Benchmarking

 Uncertainty Estimation
Confidence intervals or probabilistic outputs inform on the reliability of predictions

 Baseline Comparison
Deep learning models like CNNs and LSTMs are compared with simpler baselines
(e.g., linear regression, Random Forest) to assess added value .

5. End-to-End & Deployment Testing

 Pipeline validation
Tests ensure accuracy a cross the entire pipeline: data ingestion → preprocessing →
prediction → output delivery .

 Error analysis
Examines where models underperform—e.g., extreme weather seasons or new crop

Fig 5. 1 Stages of Testing

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 36

 Agribot insights data-driven crop identification system

The above (fig 5.1) Graphical User Interface (GUI) testing presents unique challenges
due to the vast interaction possibilities (e.g., button clicks, dialog navigation) and
complex visual elements involved. Traditional test generation methods struggle with
state-space explosion and sequence combinatorics, prompting the development of
diverse automated strategies including model-based, image- or vision-driven, planning-
based, evolutionary, and specification-driven techniques. Model-based approaches
such as GUI ripping or hierarchical planning extract structured representations of
window states and event flows to derive test sequences and associate oracles
automatically—effectively capturing user-focused behavior paths. Vision-based
methods leverage computer vision to interpret screenshots, detect UI components, and
validate GUI states beyond code-level interaction—yielding higher coverage and
adaptability . Evolutionary algorithms, seen in tools like Evo GUI Test, employ
heuristic-driven exploration to generate rich interaction scripts without requiring source
code access Specification-based frameworks (e.g., GUI Cop) use user-defined GUI
behavior contracts and trace analysis to verify GUI rendering and interactions

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 37

 Agribot insights data-driven crop identification system

5.3 Test Case Result

Table no 5.2 Test case

Test CaseID Test Case Test CaseI/P Actual Result Expectd Testcase
Result criteria
(P/F)
001 Store Xml File Xml file XMLfile store Error P
Should
come
002 Parsethe parsing Fileget parse Accepted P
xmlfile for
conversion

003 Attribute Check Identify Accepted P

identificatin individu Attribute
al s
Attribut
e
004 Weight Check Weight Analyze Accepted P
Analysi Weight
s of
individul
Attribute

005 Tree formation Formthem Formation Accepted P

- Tree

006 Cluster Check Should Accepted P

Evaluatio Evaluatio check
n n Cluster
007 Algorithm Check Should Accepted P
Performanc Evaluatio work
e n Algorithm
Properly
008 Query Check Should Accepted P
Formatio Query scheck
n Correction Query

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 38

 Agribot insights data-driven crop identification system

CHAPTER 6 . RESULTS AND DISCUSSION

6.1 Introduction

This chapter presents the results obtained from the implementation of the proposed
hybrid medical chatbot system. The performance of the CNN-based image classifier
and the agents was evaluated using appropriate quantitative and qualitative
measures. Additionally, dataset characteristics, system outputs, and overall behavior
of the chatbot are analyzed to validate the effectiveness of the system.

6. 2 Evaluation matrix

The CNN model was evaluated using standard classification metrics:

 Accuracy: 85.41%

 Precision: 84.87%

 Recall: 84.24%

 F₁-Score: 88.91%

Table no 6.1 Evaluation matrix

Algorithm Accuracy Precision Recall F₁-Score

Decision Tree 0.868 0.850 0.865 0.854

Naïve Bayes 0.937 0.941 0.941 0.941

SVM 0.930 0.926 0.939 0.930

Logistic Reg. 0.901 0.903 0.905 0.901

Random Forest 0.914 0.917 0.922 0.917

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 39

 Agribot insights data-driven crop identification system

6.3 CNN model performance

The CNN was trained for 500 epochs and achieved the following results:
• Final Training Accuracy: 98.38%
• Final Validation Accuracy: 99.88%
• Final Validation Loss: 0.0033
• Test Accuracy: 99.53%

Fig 6.1 Parallel CNN accuracy loss and profit plot

The above (fig 6.1) Parallel Convolutional Neural Network (CNN) architectures are
widely adopted to accelerate training across large datasets and complex models. These
architectures commonly employ data parallelism, where the input data is partitioned
across workers (e.g. GPUs or CPU cores), and model parallelism, where model sub-
components are trained separately and then aggregated.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 40

 Agribot insights data-driven crop identification system

Fig 6.2 Confusion matrix of parallel CNN model

The above (fig 6.2) Evaluating the performance of agricultural chatbots requires
detailed assessment metrics beyond simple accuracy. In this study, we introduce a
confusion matrix–based evaluation framework to analyze classification outcomes—
such as correctly identified farmer intents versus misclassifications. The confusion
matrix, structured into True Positives (TP), True Negatives (TN), False Positives
(FP), and False Negatives (FN), enables calculation of key performance measures:

 Accuracy = (TP +TN) ⁄ (TP +TN + FP + FN) –overall correctness.

 Precision = TP ⁄ (TP + FP) – reliability of positive predictions.
 Recall= TP ⁄ (TP + FN) – completeness in capturing actual positive intents.
 F₁-Score = 2 ×(Precision × Recall) ⁄(Precision + Recall) – balance between the two.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 41

 Agribot insights data-driven crop identification system

This approach has been applied to a crop-advisory chatbot, achieving ~95%

accuracy and F₁-score across major query categories Through detailed analysis of
FP and FN rates, the confusion matrix reveals where the bot confuses similar tasks—
such as irrigation advice vs pest alerts—and guides targeted improvements. The
methodology supports iterative refinement, particularly in handling domain- specific
language, intent overlap, and class imbalance, ensuring robust and user-centric
AI assistance for farmers.

Fig 6. 3 SVM Algorithm accuracy of Crop prediction

The above (fig 6.3 ) Support Vector Machines (SVMs) have become indispensable
in agricultural yield forecasting, particularly for regression-based crop prediction
tasks. Operating on the principle of maximizing the margin between data points
and a predictive hyperplane, SVMs utilize kernel functions—such as Radial Basis
Function (RBF) or polynomial—to effectively handle complex, non-linear
relationships found in agronomic data Despite its proven effectiveness in capturing
non-linear agronomic patterns and avoiding overfitting—especially in small to
medium datasets—SVMs are not without challenges. They can become
computationally intensive with enlarging datasets and require careful
hyperparameter tuning (e.g., C, γ) to maintain generalizability

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 42

 Agribot insights data-driven crop identification system

6.4 System output example

Fig 6.4 login page of chatbot and crop prediction

The (fig 6.4) Secure and user-friendly access is essential for digital agriculture
platforms offering personalized crop forecasting. This study presents the design
and implementation of a login interface for a crop-prediction portal that
authenticates farmers, advisors, and stakeholders before granting access to
location-specific yield forecasts, historical data dashboards, and advisory tools.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 43

 Agribot insights data-driven crop identification system

Fig 6.5 Interface of Result chatbot

The above (fig 6.5) Agriculture chatbots are intelligent virtual assistants designed to
support farmers by providing timely, accurate, and personalized agricultural
information. These systems leverage Natural Language Processing (NLP) to
understand and respond to farmer queries through text or voice interfaces, often in
regional languages. The chatbot processes user input to identify intent and extract
relevant data such as crop names, locations, or symptoms. Based on this
understanding, it interacts with integrated databases and external APIs to fetch
information on crop recommendations, weather forecasts, market prices, pest control,
and government schemes. Advanced versions incorporate artificial intelligence to
enable image-based disease detection and predictive analytics. If the chatbot
encounters complex queries beyond its scope, it escalates them to human experts

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 44

 Agribot insights data-driven crop identification system

Fig 6.6 Result of crop prediction

The above (fig 6.6) demonstrates agriculture chatbot demonstrated strong

potential in assisting farmers with real- time, personalized advice. The
combination of NLP, multilingual support, and integration with APIs resulted
in high usability and satisfaction among users. Future enhancements will focus
on expanding language options, improving image-based disease detection,
and integrating AI- based crop

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 45

 Agribot insights data-driven crop identification system

Fig 6.7 Interface of svm prediction and training

The above fig (6.7) Agribot is a web-based AI chatbot tailored for farmers, integrating
a secure user authentication interface that enables personalized access and retention of
user preferences and history. Upon login, the system offers a multilingual interaction
layer (supporting ~10–11 Indian languages), including speech-to-text and text-to-
speech capabilities, enhancing accessibility for users with varying literacy and
linguistic backgrounds The core of Agribot’s predictive capabilities lies in its
integration of supervised machine learning models. A Support Vector Machine
(SVM) classifier analyzes multi-dimensional soil and environmental features—such
as pH, nutrient levels, moisture, temperature, and localized climate—to generate crop
recommendations optimized for geographical zone and soil health. In parallel, image
inputs from users—captured via uploads—are processed using deep learning
frameworks (e.g., ResNet or similar CNNs) for plant disease detection, delivering
instant diagnostic results and management advice

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 46

 Agribot insights data-driven crop identification system

Discussions and planning

It assumes that the mitigation activity failed and the risk is a reality. Task is
done by Project manager when risk becomes reality and causes severe problems. If
the project manager effectively uses project mitigation to remove risks successfully
then it is easier to manage the risks. This shows that response that will be taken for
each risk by a manager. The main objective of the risk management plan is the risk
register. This risk register describes and focuses on the predicted threats to a
software project.
Technical Risks
Risks
 Hardware failures (motors, sensors, cameras)
 Software bugs in navigation, crop detection, or decision systems
 Poor performance in changing weather/light conditions

Management
 Use robust, weather proof components (IP-rated)

 Perform regular calibration and maintenance

 Implement redundancy in critical components (e.g., dual sensors)
 Real-time monitoring and fail-safe shut down mechanisms

 Use edge computing for local fall back if connectivity drops.

Data-Related Risks
Risks
 Inaccurate or incomplete data from sensors or AI misclassification

 Data bias (models trained only on certain crops or regions)

 Privacy concerns from collecting farm

Management
 Ensure diverse training datasets
 Regular model validation and retraining using local data
 Use data anonymization and comply with privacy laws (GDPR, etc.)

 Allow farmers to control and view collected data

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 47

 Agribot insights data-driven crop identification system

CONCLUSION AND FUTURE SCOPE

Conclusion
Provides recommendations on crop selection , planting schedules , and
nutrient management Identifies pests and diseases through image analysis and
offers solutions. Integrates with IoT sensors to monitor soil moisture and optimize
irrigation schedules. Reduces water wastage and ensures efficient water use in
farming . Offers precise recommendations for fertilizer application based on crop
needs. Minimizes environmental impact while improving productivity .

Provides real-time updates on market prices and demand. Connects farmers

with buyers, reducing dependency on middlemen. Offers tools for accessing
microloans , crop insurance and subsidy tracking. Improves financial stability and
investment capacity for farmers .Delivers real-time weather updates and alerts for
adverse conditions. Predicts risks such as pest outbreaks, droughts, or floods,
enabling timely action . Acts as a virtual platform for learning sustainable and
modern farming techniques. Provides access to tutorials, videos, and expert advice
for skill enhancement. Facilitates discussions and knowledge sharing among
farmers. Promotes a collaborative ecosystem for exchanging best practices and
innovations .Tracks produce from farm to market for better supply chain
transparency. Reduces post- harvest losses by streamlining logistics and
distribution Updates farmers on government schemes, subsidies, and agricultural
policies. Ensures compliance with regulations and maximizes access to benefits.

Future Scope
 Smart Advisory Services
 Crop selection & planning: Chatbots can suggest the best crops to grow based on
soil type, climate, and market demand. Pest and disease diagnosis: By integrating
image recognition, chatbots can identify plant diseases from photos and suggest
remedies .Fertilizer & irrigation recommendations: Personalized suggestions based on
crop type, growth stage, and soil health.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 48

 Agribot insights data-driven crop identification system

 Autonomous Field Operations: Tasks such as sowing, weeding, harvesting, and

spraying can be automated, reducing labor costs and increasing operational
efficiency.

 Livestock Management: Robots can assist in monitoring animal health, feeding, and
herding, ensuring better livestock welfare and farm management.

 Greenhouse Automation: In controlled environments, agribots can regulate

temperature, humidity, and lighting, optimizing plant growth conditions.

 Aerial Surveillance and Spraying: Drones can perform aerial surveys for crop
monitoring and precision spraying, minimizing chemical usage and environmental
impact.

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 49

 Agribot insights data-driven crop identification system

REFERENCES
1. D.katiyaret al ―AL-Powered agriculture chatbot for farmers ‖, International Research Journal
of Engineering and Technology (IRJET)- e-ISSN: 2395-0056, pISSN:2395-0072, Volume: 05
Issue: 06 — June- 2024
2. V.Himabinduet al , ―Agriculture helper chatbot ‖, 2023 International Conference on Signal
Processing and Communication (ICSC) Volume: 05 — DEC- 2023
3. K.Mandalet al ., ―Farm Friendly chatbot ‖, 2023 International Conference
on (ICIIECS) at Coimbatore Volume: 3 — March
4. Abbas iet al―The digitization of agricultural industry‖, International Research
Journal of Engineering and Technology (IRJET) Volume: 05 Issue: 02 — Feb-
2022.
5. Adkeet al.―Deep learning for cotton boll analysis , ―Sugarcane‖ published in
International Journal of Intelligent Systems and Applications Volume: 3 — August
2019.
6. Agarwalet el, ―Knowledge graph training ‖, Published by IOP Publishing Ltd Volume: 05
— OCT -2021.
7. Ahlers,D, ―Accuracy of geo names data‖ ―,published in IJSRCSEIT. UGC JournalNo
: 64718‖ — March-2013.
8. Akivaet al, ―Cranberr segmentation‖ Published in International Journal of Engineering
and Advanced Technology (IJEAT) Volume- 8 — August 2020.
9. Alam, a el―AI in education ‖ IJARCSSE,vol. 6, Issue 5, 2021
10. Katherine el, ―Agribot ‖ IJRITCC, vol. 2, Issue 2,292- 297,2018.
11. Atony parr (2019). "Farmchat (2019) " International Journal of
Computer Applications [DOI / Search on IEEE Xplore or ResearchGate]
12. Krishnswami(2020). "Krishibot " International Journal of Scientific &
Engineering Research [Available on: www.ijser.org]
FAO&ITU(2022). "AI in Agriculture: The Future of Smart Farming inDeveloping

13. AgAsk: An Agent to Help Answer Farmer's Questions From Scientific Documents
– Bevan Koopman et al.
14. AgroLLM: Connecting Farmers and AgriculturalPracticesthrough Large Language
Models – Dinesh Jackson Samuel et al.
15. AgriBot: Your Intelligent Farm Assistant – Apurva Dipak Jadhavet al.

16. Smart Chatbot for Farmers’ Agricultural Guidance and Assistance – K. Anusha et al.

17. Norm: AI Agriculture Advisor byFarmers Business Network – Farmers Business

Network

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 50

 Agribot insights data-driven crop identification system

18. Patel, H., & Gamit, A. (2019)."Crop prediction using machine learning." International
Research Journal of Engineering and Technology (IRJET), 6(3), 1638– 1641.
https://www.irjet.net/archives/V6/i3/IRJET- V6I3322.pdf
19.Singh, P., & Arora, A. (2020)."Crop yield prediction using machine learning algorithms:
Are view." Materials Today: Proceedings, Elsevier.
DOI: 10.1016/j.matpr.2020.10.813
20.Kaur, H., &Singh, M. (2018)."A review on crop yield prediction using machine
learning techniques. "International Journal of Computer Applications, 179(18), 7–
10. https://doi.org/10.5120/ijca2018917201

Department of CSE, Nutan College of Engineering and Research, Talegaon , pune 51