Madhav Institute of Technology & Science, Gwalior
(Deemed to be University)
(Declared under Distinct Category by Ministry of Education, Government of India)
NAAC Accredited with A++ Grade
Skill Based Mini Project
Report on
Prediction on Song Popularity
Submitted By:
Aryan Dubey
0901CD211012
Submitted To:
Prof. Mahesh Parmar
Assistant Professor
DEPARTMENT OF COMPUTER SCIENCE & ENGINEERING
Madhav Institute of Technology & Science, Gwalior
(Deemed to be University)
(Declared under Distinct Category by Ministry of Education, Government of India)
NAAC Accredited with A++ Grade
January-June 2024
1
Madhav Institute of Technology & Science, Gwalior
(Deemed to be University)
(Declared under Distinct Category by Ministry of Education, Government of India)
NAAC Accredited with A++ Grade
CERTIFICATE
This is certified that Aryan Dubey (0901CD211012) has submitted the project report titled
“Handwritten Digit Recognition” under the mentorship of Prof. Mahesh Parmar,
Assistant Professor in partial fulfilment of the requirement for the award of degree of
Bachelorof Technology in Computer Science and Engineering from Madhav Institute of
Technology and Science, Gwalior.
Prof. Mahesh Parmar
Faculty Mentor
Assistant Professor
Computer Science and Engineering
2
Madhav Institute of Technology & Science, Gwalior
(Deemed to be University)
(Declared under Distinct Category by Ministry of Education, Government of India)
NAAC Accredited with A++ Grade
DECLARATION
I hereby declare that the work being presented in this project report, for the partial fulfilment of
requirement for the award of the degree of Bachelor of Technology in Computer Science and
Engineering at Madhav Institute of Technology & Science, Gwalior is an authenticated and
original record of my work under the mentorship of Prof. Mahesh Parmar , Assistant
Professor, Computer Science and Engineering.
I declare that I have not submitted the matter embodied in this report for the award of any degree
or diploma anywhere else.
Aryan Dubey
0901CD211012
3rd Year,
Computer Science and Design
3
Madhav Institute of Technology & Science, Gwalior
(Deemed to be University)
(Declared under Distinct Category by Ministry of Education, Government of India)
NAAC Accredited with A++ Grade
ACKNOWLEDGEMENT
The full semester project has proved to be pivotal to my career. I am thankful to my institute,Madhav
Institute of Technology & Science Gwalior to allow me to continue my disciplinary/interdisciplinary
project as a curriculum requirement, under the provisions of theFlexible Curriculum Scheme (based on the
AICTE Model Curriculum 2018), approved by the Academic Council of the institute. I extend my
gratitude to the Director of the institute, Dr. R. K. Pandit and Dean Academics, Dr. Manjaree Pandit
for this.
I would sincerely like to thank my department, Department of Computer Science and Engineering,
for allowing me to explore this project. I humbly thank Dr. Manish Dixit, Professor and Head,
Department of Computer Science and Engineering, for his continued support during the course of this
engagement, which eased the process and formalities involved.
I am sincerely thankful to my faculty mentors. I am grateful to the guidance of Prof. Mahesh Parmar,
Computer Science and Engineering, for her continued support and guidance throughout the project. I am
also very thankful to thefaculty and staff of the department.
Aryan Dubey
0901CD211012
3rd Year,
Computer Science and Design
4
Table of content
1. Micro Project: Training of Data 6
1.1 Introduction 6
1.2 Objectives 7
1.3 Implementation 7-8
2. Macro Project:Data Visualization 9
2.1 Introduction 9
2.2 Objectives 9
2.3 Implementation 10-11
3. Skill-based mini-project 12
3.1 Introduction 12
3.2 Objective 12
3.3 Implementation 12
3.4 Code & Output 13-15
4. Result 16
5. Conclusion 17
5
MICRO PROJECT
Training of Data
Introduction
Training data is a crucial component in machine learning and artificial intelligence. It refers
to the dataset that is used to train a machine-learning model. In supervised learning, which
seems to be the case here as you're predicting digits, training data typically consists of input-
output pairs. Each input corresponds to a set of features or attributes, and each output
corresponds to the label or target variable we want the model to predict.
Objective
1. Quality: The quality of training data significantly impacts the performance of the model.
High-quality data should be accurate, representative of the real-world scenario, and free
from errors or biases.
2. Quantity: The amount of training data available can also influence model
performance. Generally, more data allows the model to learn better patterns and
generalize well to unseen example.
3. Features and Labels: Training data consists of both input features and corresponding labels.
Features are the attributes or characteristics of the data that the model learns from, while labels are
the target outputs the model aims to predict..
4. Preprocessing: Before training a model, it's often necessary to preprocess the training
data. This may involve tasks such as normalization, feature scaling, handling missing
values, and encoding categorical variables..
5. Balancing: In classification tasks, it's essential to ensure that the training data is
balanced across different classes to prevent the model from being biased towards the
majority class.
6
Implementation
7
8
Macro Project
Data Visualization
Introduction
Data visualization plays a crucial role in understanding, analyzing, and presenting the data used
in digit recognition tasks. In digit recognition, data visualization techniques help in exploring the
characteristics of the input images, understanding the distribution of the data, and interpreting the
model's predictions.
Objective
1. Exploratory Analysis: Data visualization allows researchers and practitioners to
explore the dataset before training the model. This includes visualizing sample
images, examining class distributions, and identifying potential challenges such as
class imbalances or noisy data.
2. Model Evaluation: After training the digit recognition model, data visualization
techniques are used to evaluate its performance. Visualization tools help in analyzing
the model's predictions, understanding its strengths and weaknesses, and identifying
areas for improvement.
3. Interpretability: Visualization aids in interpreting the predictions made by the
model. For example, visualization techniques such as heatmaps can highlight
the regions of the input image that contributed most to the model's decision.
4. Comparison: Data visualization facilitates the comparison of different models or
approaches. By visualizing the performance metrics or prediction results side by side,
stakeholders can make informed decisions about which model to deploy.
9
Implementation
10
11
Skill Based Mini Project
Introduction
Handwritten digit recognition is a foundational problem in machine learning and computer vision, with
diverse applications spanning optical character recognition (OCR), automated mail sorting, and medical
imaging analysis. In this project, we aim to employ convolutional neural networks (CNNs), a subtype of
deep learning models renowned for their efficacy in image classification tasks, to tackle handwritten digit
recognition using the MNIST dataset. Leveraging TensorFlow and Keras, we will preprocess the data,
construct a CNN architecture, and train it on the dataset to achieve accurate digit classification. This
endeavor serves to exemplify the utility of deep learning methodologies in real-world image classification
scenarios.
Objectives
1. Implement a handwritten digit recognition system utilizing convolutional neural networks (CNNs).
2. Preprocess the MNIST dataset, including normalization of pixel values and reshaping images for
input into the CNN.
3. Design and construct a CNN architecture using TensorFlow and Keras to effectively learn and
classify handwritten digits.
4. Train the CNN model on the training dataset to optimize its performance in digit classification.
5. Evaluate the trained model on a separate test dataset to measure its accuracy and generalization
ability.
6. Explore the impact of different CNN architectures, hyperparameters, and optimization techniques
on the performance of the digit recognition system.
7. Provide insights into the efficacy of deep learning approaches for solving handwritten digit
recognition tasks.
8. Demonstrate the practical applicability of CNNs in image classification problems beyond
handwritten digit recognition.
Implementation
1. Dataset Loading: Utilize TensorFlow's mnist.load_data() function to load the MNIST dataset,
consisting of handwritten digit images and their corresponding labels.
2. Data Preprocessing: Normalize the pixel values of the images to a range between 0 and 1 by
dividing by 255. Reshape the images to have a single channel and a size of 28x28 pixels to conform
to the input requirements of the CNN.
3. Visualization: Use matplotlib to visualize a subset of the training images along with their
corresponding labels to gain insights into the dataset.
4. Model Construction: Design a CNN architecture using TensorFlow's Keras API. Define the
layers including convolutional layers, pooling layers, and fully connected layers. Experiment with
different architectures to optimize performance.
5. Model Compilation: Compile the CNN model, specifying the loss function, optimizer, and
evaluation metrics to be used during training.
6. Model Training: Train the CNN model on the preprocessed training dataset using the model.fit()
function. Adjust the number of epochs and batch size as needed to optimize training performance.
7. Model Evaluation: Evaluate the trained model on the separate test dataset using the
model.evaluate() function to measure its accuracy and performance in digit classification.
12
8. Hyperparameter Tuning: Experiment with different hyperparameters such as learning rate,
dropout rate, and kernel size to fine-tune the model and improve its performance.
9. Visualization of Results: Visualize the model's predictions on the test dataset to analyze its
performance and identify any misclassifications or areas for improvement.
10. Deployment: Deploy the trained model for real-world applications, such as integrating it into a
web or mobile application for real-time handwritten digit recognition.
Code & Output
13
14
15
Result
1. Achieved an accuracy of approximately 99% on the test dataset after training the CNN model on
the MNIST dataset.
2. Visualized the model's predictions on the test dataset, revealing robust performance in accurately
recognizing handwritten digits.
3. Identified occasional misclassifications, particularly for ambiguous or poorly written digits,
highlighting areas for potential improvement.
4. Demonstrated the effectiveness of the CNN architecture in solving handwritten digit recognition
tasks, showcasing the power of deep learning.
5. Successfully deployed the trained model for real-world applications, enabling integration into OCR
systems, automated mail sorting machines, or mobile applications.
6. Highlighted the potential impact of deep learning methodologies, particularly CNNs, in addressing
diverse image classification challenges across various domains.
16
Conclusion
In summary, the project has substantiated the efficacy of convolutional neural networks (CNNs) in the
domain of handwritten digit recognition, specifically leveraging the MNIST dataset. Employing
TensorFlow and Keras, a robust CNN architecture was meticulously engineered, achieving a commendable
accuracy threshold of approximately 99% on the test dataset.
Through meticulous visualization and analysis of the model's predictions, it has been demonstrated that the
CNN exhibits consistent and proficient recognition of handwritten digits, showcasing its adeptness in
generalizing across diverse writing styles and variations. Although occasional misclassifications were
observed, particularly in instances of ambiguous digits, the overarching performance remained highly
satisfactory.
The successful deployment of the trained model underscores its tangible applicability in real-world
scenarios, spanning optical character recognition (OCR) systems, automated mail sorting infrastructures,
and mobile applications necessitating adept digit recognition capabilities.
This project stands as a testament to the transformative potential of deep learning paradigms, notably CNNs,
in addressing intricate image classification challenges. By virtue of continual refinement and optimization,
such models hold promise for catalyzing advancements in artificial intelligence and machine learning,
thereby fostering innovation across a myriad of domains.
17
REFERENCES
[1] M. Dyrmann, H. Karstoft, and H. S. Midtiby, “Plant species classification using deep convolutional neural
network,” Biosyst. Eng., vol. 151, no. 2005, pp. 72–80, 2016.
[2] S. S. Yadav and S. M. Jadhav, “Deep convolutional neural network based medical image classification for disease
diagnosis,” J. Big Data, vol. 6, no. 1, 2019.
[3] N. A. Mahmon and N. Ya’Acob, “A review on classification of satellite image using Artificial Neural Network
(ANN),” Proc. - 2014 5th IEEE Control Syst. Grad. Res. Colloquium, ICSGRC 2014, pp. 153–157, 2014
.
[4] T. Guo, J. Dong, H. Li, and Y. Gao, “Simple convolutional neural network on image classification,” 2017 IEEE
2nd Int. Conf. Big Data Anal. ICBDA 2017, pp. 721–724, 2017.
[5] S. Kaymak, A. Helwan, and D. Uzun, “Breast cancer image classification using artificial neural networks,”
Procedia Comput. Sci., vol. 120, pp. 126–131, 2017.
[6] A. Verma, S. Shekhar, and H. Garg, “Plant Disease Classification Using Deep Learning Framework,” Proc. Int.
Conf. Comput. Intell. Sustain. Eng. Solut. CISES 2022, no. May, pp. 512– 518, 2022.
[7] S. R. Maniyath et al., “Plant disease detection using machine learning,” Proc. - 2018 Int. Conf. Des. Innov. 3Cs
Comput. Commun. Control. ICDI3C 2018, pp. 41–45, 2018.
[8] L. Li, S. Zhang, and B. Wang, “Plant Disease Detection and Classification by Deep Learning - A Review,” IEEE
Access, vol. 9, no. Ccv, pp. 56683–56698, 2021.
[9] M. R. Ullah, N. A. Dola, A. Sattar, and A. Hasnat, “Plant Diseases Recognition Using Machine Learning,” Proc.
2019 8th Int. Conf. Syst. Model. Adv. Res. Trends, SMART 2019, pp. 67–73, 2020.
18
