What is Classification?

Classification is a type of supervised learning where the AI model sorts data into
predefined categories or classes.
Example: Email spam detection – The model analyzes emails and classifies
them as either "Spam" or "Not Spam" based on patterns in the text.
What is Regression?
Regression is a type of supervised learning where the AI model predicts
continuous numerical values based on input data.
Example: House price prediction – The model estimates the price of a house
based on factors like size, location, and number of rooms.
Types of AI
AI is classified into three types based on its capabilities:
1. Weak AI (Narrow AI) – Designed for a specific task, such as virtual
assistants (e.g., Siri, Alexa).
2. Strong AI (General AI) – Has human-like intelligence and can perform
any intellectual task. It does not yet exist.
3. Super AI – Hypothetical AI that surpasses human intelligence and can
make its own decisions.
AI Integration with Non-AI Machines (Example)
AI can be integrated into non-AI machines by adding smart features. For
example, a normal CCTV camera records footage, but when integrated with AI,
it can recognize faces, detect movements, and alert security in real-time.
How AI Provides a Personalized Online Experience
AI tracks user preferences and behaviors to provide tailored recommendations.
For example:
 Streaming platforms (Netflix, YouTube) suggest content based on
watch history.
 E-commerce sites (Amazon) recommend products based on previous
purchases.
 Social media (Instagram, Facebook) shows posts and ads that match
user interests.
Applications of Data Science, NLP, and Computer Vision
 Data Science – Used in fraud detection, stock market analysis, and
healthcare predictions.
 Natural Language Processing (NLP) – Used in chatbots, voice
assistants, and language translation (e.g., Google Translate).
 Computer Vision – Used in facial recognition, medical imaging, and self-
driving cars.
Relation Between the Three AI Domains
The three domains of AI—Data Science, NLP, and Computer Vision—work
together to improve AI applications. For example, a voice assistant like Siri
uses NLP to understand speech, Data Science to process user data, and
Computer Vision to analyze images (if needed).

AI Project Cycle
The process of converting a real-life problem into an AI-based model is known as
the AI project cycle. It involves the following stages:

1. Problem Scoping
The first step of the AI project cycle involves analyzing and understanding the
problem with the help of the 4W Canvas:
 Who – The people directly or indirectly affected by the issue.
 What – The problem faced; analyzing the situation deeply.
 Where – A broader look into the problem by identifying areas where it is
most prominent.
 Why – How the solution benefits society and stakeholders.

2. Data Acquisition
 Data refers to raw figures, numbers, facts, and statistics.
 Acquisition refers to the process of collecting data for machine learning.
 Sources of Data Acquisition:
o Sensors – Convert real-world phenomena like climate, temperature,
and current signals into inputs.
o Cameras – Obtain visual data through webcams and CCTV.

o Surveys – Provide quantitative and qualitative data directly from

consumers.
o API (Application Program Interface) – Pieces of code that
interconnect different applications.
o Web Scraping – Extracts information from websites using a web
browser and arranges it into CSVs and spreadsheets.

3. Data Exploration
  The process of arranging collected data into charts, graphs, and other
visual representations.
 Helps extract useful information for analysis.
 Example: Google Charts.

4. Modelling
Developing algorithms (models) that produce intelligent output based on
trained data.
Types of Approaches
1. Rule-Based Approach
o The developer feeds the model with data along with a set of ground
rules.
o Used when a labeled or known dataset is provided.

2. Learning-Based Approach
o Used when an unknown or unlabeled dataset is provided.

o The model is trained with data and answers to develop algorithms

matching given data.
Types of Learning-Based Approaches
 Supervised Learning
o The model is fed a labeled dataset where input and corresponding
output data are provided.
o Example: Email spam detection, Image recognition.

 Unsupervised Learning
o The model is fed unlabeled data to identify intrinsic structures and
hidden patterns.
o Example: Natural Language Processing (NLP), Fraud
Detection, Data Mining.
 Reinforcement Learning
o The model (agent) makes decisions by taking actions in an
environment to maximize cumulative rewards.
o Example: Gaming, Robotics, Autonomous Vehicles.

5. Evaluation
The evaluation process is based on the output received from a trained model and
cross-checking it with real answers.
Key Evaluation Metrics
 True Positive (TP) – The number of positive examples correctly
classified as positive.
 True Negative (TN) – The number of negative examples correctly
classified as negative.
 False Positive (FP) – The number of negative examples incorrectly
classified as positive.
 False Negative (FN) – The number of positive examples incorrectly
classified as negative.
Performance Measures
 Prediction – The answer given by the trained model.
 Reality – The actual outcome after the prediction is made.
 Accuracy – The percentage of correct predictions out of all observations.
 Precision – The percentage of true positives out of all cases where the
predictions were positive.
 Recall – The fraction of positive cases that were correctly identified.
 F1 Score – A measure of the balance between precision and recall.

6. Advantages and Disadvantages of Artificial Neural Networks (ANN)

Advantages of ANN
1. Ability to learn complex patterns – ANNs can identify intricate
relationships in data that traditional algorithms may miss.
2. Self-learning capability – Once trained, ANNs can recognize patterns
and make decisions without human intervention.
3. Handling of large datasets – ANNs are efficient in processing vast
amounts of data, making them ideal for big data applications.
4. Fault tolerance – ANNs can still function properly even when some
neurons fail, ensuring system reliability.
Disadvantages of ANN
1. Black-box nature – ANNs lack transparency, making it difficult to
interpret how they arrive at decisions.
2. Computationally expensive – Training neural networks requires
significant computational power and time.
3. Requirement for large datasets – ANNs need vast amounts of labeled
data for effective learning, which can be difficult to obtain.
4. Overfitting risk – ANNs may memorize training data instead of
generalizing, leading to poor performance on new data.