CS Academy , Erode

Unit 2. Advanced concepts of Modeling in AI

1. What is an AI model?
An AI model is a program trained to recognize patterns using a dataset. It is created by
designing algorithms that can learn from data to produce intelligent outputs.
2. Differentiate between Rule-Based and Learning-Based AI models.
 Rule-Based: The developer sets fixed rules. The model doesn’t learn or improve
from new data after training.
 Learning-Based: The model learns from data and can adapt when the input data
changes.
3. What is supervised learning?
In supervised learning, the dataset is labeled, meaning the input data is paired with correct
output. The model learns the relationship between them to make predictions.
4. What is classification in AI?
Classification is a type of supervised learning where the data is discrete (non-continuous).
For example, classifying emails as “spam” or “not spam.”
5. What is regression in AI?
Regression is a supervised learning method used with continuous data to predict
numerical values—like predicting future salary or stock prices.
6. Differentiate between classification and regression.
 Classification: Works with discrete labeled data (e.g., pass/fail).
 Regression: Works with continuous labeled data (e.g., salary prediction).
7. What is unsupervised learning?
Unsupervised learning uses unlabeled datasets. The model finds patterns or structures in
the data without knowing the outcome in advance.
8. What is clustering?
Clustering is an unsupervised learning method where similar data points are grouped
based on patterns. It's used in recommendation systems.
9. Explain Dimensionality Reduction with an example.
It reduces the number of variables in data while preserving meaningful information. E.g.,
reducing a 3D ball image to a 2D photo loses depth but keeps useful visual information.
10. What is Association Rule in AI?
It’s an unsupervised technique that finds relationships between variables. E.g., “If a
customer buys bread, they likely buy butter.”
11. What is Reinforcement Learning (RL)?
RL is where software learns by interacting with its environment using trial and error. It
learns optimal behavior through feedback from its actions.
12. What is Deep Learning?
A type of machine learning where software trains itself using large datasets to learn
patterns and make decisions without human intervention.
13. Name two types of Deep Learning models.
 Artificial Neural Networks (ANN)
 Convolutional Neural Networks (CNN)
14. What is an Artificial Neural Network (ANN)?
ANNs are inspired by the human brain. They can identify patterns in large datasets and are
used in tasks like speech recognition and forecasting.
15. What is a Convolutional Neural Network (CNN)?
CNNs are deep learning models used for image and video recognition. They analyze visual
features to recognize objects, faces, etc.
16. Define Neural Network.
A neural network is a system inspired by the human brain, used to recognize patterns in
large data. It solves problems like facial recognition and language translation.
17. What are the layers of a Neural Network?
 Input Layer: Collects raw data.
 Hidden Layers: Process data using weights and biases.
 Output Layer: Produces the result.
18. What is a weight in a neural network?
A weight is a number that shows how important an input is. Inputs are multiplied by
weights; larger weights mean more importance.
19. What is a bias in neural networks?
Bias is a number added after multiplying the input by weight. It helps the model make
better predictions by shifting the output.
20. Explain the park decision-making example using neural networks.
Inputs like weather and umbrella are converted to numbers, multiplied by weights, and
added with bias. If the result > 0, the model says "Go to park," else "Stay home."
21. Why is rule-based AI called "static learning"?
Because once the rules are set and the model is trained, it cannot learn from new data or
feedback. The model always follows the same instructions without adaptation.
22. Why are labels important in supervised learning?
Labels act as tags that help the model learn the correct output during training. Without
them, the model won’t know what it's trying to predict.
23. Give an example of supervised learning in daily life.
Grading students based on marks is an example—grades (labels) are assigned to numerical
marks (data).
24. Give an example of unsupervised learning in daily life.
Grouping similar movies on a streaming platform based on viewing patterns without
knowing users’ preferences.
25. What does "continuous data" mean in regression?
It means data that can take any value in a range. For example, temperature, salary, and
age are all continuous data types.
26. How does dimensionality reduction affect data quality?
Dimensionality reduction means reducing the number of features or inputs in a dataset.
 It can remove some information, especially the less important details.
 But it helps to keep the most useful parts of the data.
 This makes the data easier and faster for the computer to understand and process.
👉 Example:
Imagine holding a 3D ball. If you take a photo of it, the image becomes 2D—you can’t see
the back side. Some information is lost, but the photo still gives a good idea of what the
ball looks like.
27. What are common uses of Association Rule learning?
Used in market basket analysis, like recommending products that are often bought
together on e-commerce sites.
28. How is reinforcement learning different from supervised learning?
In supervised learning, the model is trained with correct answers. In reinforcement
learning, the model learns from trial and error by receiving rewards or penalties.
Example:
 Supervised learning: A student learns math by studying solved problems.
 Reinforcement learning: A robot learns to walk by trying, falling, and improving
based on feedback.
29. What makes deep learning powerful?
It automatically learns complex patterns from massive datasets and improves over time,
making it ideal for tasks like image recognition and language translation.
30. Where are CNNs commonly used?
CNNs are mainly used in:
 Facial recognition
 Medical imaging
 Self-driving cars
 Image classification
31. Where are Artificial Neural Networks (ANNs) commonly used?
Artificial Neural Networks (ANNs) are commonly used in solving complex problems where
recognizing patterns is essential. Some typical applications include:
 Speech Recognition (e.g., voice assistants like Siri or Google Assistant)
 Financial Forecasting (e.g., predicting stock trends or credit scoring)
 Handwriting Recognition (e.g., converting handwritten notes to text)
 Medical Diagnosis (e.g., analyzing patient data for disease prediction)
 Customer Support Chatbots (e.g., automated responses in online support)

32. What is a Neural Network?

A Neural Network is a part of Artificial Intelligence that works like the human brain. It is made up of
layers with small parts called nodes or neurons. Each node does a small task and passes the result to the
next layer.

There are three main types of layers:

1. Input Layer – It takes in the data (like an image or numbers). No processing happens here.
2. Hidden Layer(s) – These layers process the data using weights and biases. This is where
learning happens. The data goes through calculations and is adjusted until the right answer is found.
3. Output Layer – This layer gives the final result to the user. Like the input layer, it doesn’t process
the data.

Neural Networks are very good at handling large data, such as images or sound. They are used in real-
life for facial recognition, chatbots, predicting prices, and more.
Differentiate Questions :
1. Differentiation between AI, ML, and DL.
AI (Artificial
🔍 Aspect Intelligence)
ML (Machine Learning) DL (Deep Learning)

Makes computers think and Helps computers learn A smart system that learns on its
What it is
act like humans from data own using big data
Follows rules or logic given Learns from examples and Uses brain-like networks to learn
How it works
by humans past experience automatically
Learning Needs human rules to make Learns patterns step by step Learns deep patterns without
style decisions from data human help
Can work with small or big Needs enough data to learn Needs a huge amount of data to
Data needs
data well perform best
Smartness Depends on what humans Gets smarter with each new Is very smart — improves itself
level teach example by learning from experience
Big concept (includes ML &
Connection A part of AI A part of ML (so also part of AI)
DL)
Example 1 A chatbot that replies to A system that learns from
A system that recognizes your face
your questions using fixed your past music choices to
to unlock your phone
rules recommend new songs
Example 2 A virtual assistant (like A spam filter that learns A self-driving car that learns to
Alexa or Siri) that follows which emails are spam recognize roads, people, and signs
commands like "Play music" based on what you mark from images
Example 3 A calculator app that follows An app that predicts your An app that reads handwritten
instructions to solve exam score from your past notes and turns them into typed
equations marks and study time text
Human Rules and logic written by ML= Data + Algorithm DL = Lots of data provided by
Help humans chosen and designed by a humans + Neural network
human. algorithm built and trained by the
system.

2. What is the difference between regression and classification in AI?

Feature Regression Classification

Type of Works with continuous data Works with discrete data
Data (numbers) (categories)
Output Predicts a numerical value Predicts a class or label
Example Predicting salary, stock prices, Classifying emails as spam or not
temperature spam
Used For Forecasting, trend analysis Grouping, sorting, tagging
Graph
3. Difference between clustering and dimensionality reduction .
Feature Clustering Dimensionality Reduction
Type of Learning Unsupervised Learning Unsupervised Learning
Purpose Group data into similar clusters Reduce the number of features in data
Output Labels or cluster groups (e.g., A, Simplified version of original dataset
B, C)
Data Helps discover hidden group Helps remove redundant or less important
Interpretation patterns data
Example Customer segmentation Image compression (3D to 2D)

4. Difference between rule based and learning based approach .

Feature Rule-Based Approach Learning-Based Approach
How it Follows fixed rules written by a Learns from data and improves over time
works programmer
Adaptability Cannot adapt to new situations Can adapt to changes in data
Learning No learning – it’s static Can learn from new examples – dynamic
Example “If temperature > 38°C → Fever” Predicting disease based on symptoms
patterns
Safari Detects elephant only if all conditions Learns flexible patterns — detects even in new
Example exactly match conditions
Example Rings bell at fixed times only Learns and adjusts ringing time based on real
student patterns
Real-Life May ring the bell even on holidays Learns there’s no school on holidays and doesn't
Result ring the bell
5. Difference Between Training and Testing Data Set

Aspect/Poin Training Data Set Testing Data Set

t
Definition A collection of examples given to the Used to test the accuracy of the model
model to analyze and learn
Analogy Like how a teacher teaches a topic using Like how a teacher takes a class test to evaluate
examples and illustrations student understanding
Labeling A set of labeled data is used to train the Test is performed without labeled data, and
AI model results are later verified with labels
Purpose To help the model learn from examples To evaluate how well the model has learned

6.Difference Between ANN and CNN .

Aspect Artificial Neural Network (ANN) Convolutional Neural Network (CNN)

Inspiration Modeled on the human brain and A type of Deep Learning algorithm
nervous system specifically designed for image-related
tasks
Feature Can automatically extract features, but Automatically extracts features from
Extraction not optimized for spatial data like images images using filters (convolutions)
Structure Made up of fully connected layers Uses convolutional layers, pooling
layers, and sometimes fully connected
layers
Input Type Works well with structured data (like Designed to work with image data
 numbers, text, etc.)
Use Case Useful for large general datasets Best suited for image recognition, object
detection, and visual tasks
Learning Each node is a machine learning Assigns learnable weights and biases to
Components algorithm different parts of the image

7 Difference Between Weight and Bias.

Weight Bias
A weight is a value that defines the importance of an A bias is an additional parameter added to adjust the
input feature (Tells how important an input is.) output (Helps the model adjust the output.)
It is multiplied with the input It is added after multiplication
Without weights, the model can't learn patterns Without bias, the model can't shift the output
properly
Eg: Weight says how much marks matter. Eg: Bias adds extra marks (like grace marks) to adjust
the result
8 Difference Between Classification and Clustering

Aspect Classification Clustering

Use of Labels Uses predefined classes (labels are No predefined labels; finds patterns on its own
given)
Purpose Assigns objects to known categories Groups similar objects into clusters based on
similarity
Type of Supervised Learning Unsupervised Learning
Learning
Example 1 Labeling fruits as "Apple", "Banana", or Grouping fruits by color/size without knowing their
"Mango" names
Example 2 Marking emails as "Spam" or "Not Grouping emails based on topics without knowing
Spam" the categories

9 Difference Between Supervised, Unsupervised, and Reinforcement Learning

Aspect Supervised Learning Unsupervised Learning Reinforcement Learning

Data Requires labeled data Uses unlabeled data Does not require large pre-existing
Requirement data

Knowledge You must know the input You know the data, but not Works even when you don’t know all
About Data and output – (You teach the output labels – (You the data or environment –(The
with answers (like giving only give questions, no student learns by doing and getting
questions and answers to answers (the student finds rewards or penalties — even if you
a student).) patterns)) don’t explain the full problem.)
Problem Type Predicting output from Grouping or pattern Learning by trial and error
known examples detection

Learning Style Learns from examples Learns by finding structure Learns by reward and punishment
system

When to Use When you have clear When exploring data When facing complex, changing
data and labels without labels problems with uncertain outcomes

Case-Based Question:
1.A bank wants to analyze customer transaction data to identify and flag suspicious or
fraudulent activities. However, the bank has not defined what constitutes a suspicious
transaction, and there are no labeled examples of "fraud" or "not fraud" available.

Based on this scenario, answer the following:

a) What type of machine learning approach is used in this case?
b) Why is this approach suitable for the given problem?

Answer:

a) Unsupervised Learning
b) Since there are no labeled examples of fraudulent transactions, the model needs to learn patterns on its
own. It identifies unusual or anomalous behavior (outliers) in the data and flags those transactions as
potentially fraudulent.

2. How does AI make a decision like whether to go out to the park or not?
Use the example below:

 I have a jacket → Yes (1)

 I don’t have an umbrella → No (0)
 It is sunny now → Yes (1)
 Weather forecast says it will rain → No (0)
 The importance (weights) for each:
o Jacket: 0.2
o Umbrella: 0.1
o Sunny now: 0.6
o Forecast: 0.1
 Bias weight: -0.5
 Rule: If the final result is more than 0, I will go out. Otherwise, I will stay in.

What will the AI decide? Will I go out or not?

Answer:
Output =(1 × 0.2) + (0 × 0.1) + (1 × 0.6) + (0 × 0.1) + (1 × -0.5)
= 0.2 + 0 + 0.6 + 0 - 0.5
= 0.3

Since 0.3 is more than 0, the AI will decide:

You will go out to the park.

3 . You are building a model to predict whether a customer is eligible for a bank loan or not.
What type of model is this – Classification or Regression?

Answer:
Classification – This is a Binary Classification problem because the model predicts a Yes or No output
(discrete values).

4. You are building a model to predict the weather for the next 24 hours.
What type of model is this – Classification or Regression?

Answer:
Regression – This is a Regression problem because the output is a continuous value (like temperature,
which changes over time).

5. Read the following cases and answer the question:

Case 1:
 A model is built to predict the temperature of a city at different times of the day.

Case 2:
A company wants to create a model to predict the price of a house based on:

 Number of bedrooms
 Carpet area
 Garage size

Case 3:
A car-selling website uses a model to predict the selling price of a used car based on:

 Fuel type
 Years of service
 Number of previous owners
 Kilometers driven
 Transmission type

What type of machine learning model is used in all the above cases? Why?

Answer:

All the above cases use a Regression Model because:

 The output in each case is a continuous numerical value (temperature, house price, car price).
 These values can vary across a range and are not limited to fixed categories.
 Regression is used to predict real numbers based on input features.

6. A model learns from past weather data (like location, temperature, humidity, and season)
and then predicts whether tomorrow will be hot or cold.What type of model is this?

Answer:

It is a Classification Model because it predicts categories (hot or cold).

Additional Questions :
✅ Case Study 1: Spam Email Filter
Scenario:
A school implemented a spam email filter using machine learning. Over time, the filter improved at
identifying spam but sometimes marked important newsletters as spam. The team used a labeled dataset
during training.

Q1: What type of learning approach was used in this case?

Answer: Supervised Learning

Q2: Why is labeled data important for such a model?

Answer: Labeled data helps the model associate input features with the correct output (spam or not spam)
during training.

Q3: How could the model be improved to reduce false positives?

Answer: By training with a more diverse and balanced dataset, using feedback loops, and optimizing the
model with precision-recall tuning.

✅ Case Study 2: E-Commerce Product Recommendation

Scenario:
An e-commerce platform uses AI to recommend products to its users. Based on purchase history, the
system groups customers and identifies likely product combinations.

Q1: Which machine learning models are likely used in this scenario?
Answer: Clustering and Association Models

Q2: Explain the roles of clustering and association models here.

Answer: Clustering groups customers with similar preferences, while association finds frequent item
combinations for recommendations.

Q3: Is labeled data required in this case? Justify your answer.

Answer: No. Both clustering and association are unsupervised learning models and do not require labeled
data.

✅ Case Study 3: Self-Driving Car Navigation

Scenario:
A tech company is training a self-driving car to recognize traffic signs and take proper actions. Initially, the
car makes errors, but with repeated trials and rewards for correct moves, it learns to perform better.

Q1: Identify the learning model used in this scenario.

Answer: Reinforcement Learning

Q2: How is this model different from supervised learning?

Answer: Reinforcement learning uses a trial-and-error method with feedback in the form of rewards,
whereas supervised learning uses labeled datasets.

Q3: What role do rewards and penalties play in the system?

Answer: Rewards reinforce correct behavior, and penalties discourage wrong actions, helping the model
learn optimal decisions.
✅ Case Study 4: Hospital Patient Risk Prediction
Scenario:
A hospital uses an AI model to predict patients at high risk based on health records. The data includes age,
blood pressure, past illnesses, etc., and is labeled with patient outcomes.

Q1: What kind of learning approach is being used?

Answer: Supervised Learning

Q2: Would classification or regression be more suitable here? Why?

Answer: Classification – because the output is categorical (e.g., high-risk or low-risk).

Q3: How can ethical concerns like bias be addressed in such a model?
Answer: By ensuring diverse training data, using fairness-aware algorithms, and applying ethical
frameworks like bioethics.

✅ Case Study 5: Music Streaming App

Scenario:
A music app tracks user behavior like favorite genres, skipped songs, and listening time. Based on this, it
suggests playlists and songs without knowing user preferences in advance.

Q1: Which type of learning model is applied here?

Answer: Unsupervised Learning

Q2: How does the system learn without labels?

Answer: It analyzes data patterns to group similar behaviors without needing predefined categories.

Q3: Suggest one benefit and one limitation of using this approach.
Answer:
Benefit: Can discover hidden trends and personalize content.
Limitation: May group users incorrectly due to ambiguous or noisy data.

✅ Case Study 6: Handwriting Recognition

Scenario:
An app is built to recognize handwritten digits (0–9). It is trained on thousands of labeled image samples of
digits and predicts new inputs based on learned features.

Q1: What type of learning and sub-model is used here?

Answer: Supervised Learning using Classification

Q2: Why is a neural network suitable for this task?

Answer: It can automatically extract features from image data and handle high-dimensional input.

Q3: How would the model’s performance be evaluated?

Answer: Using accuracy, confusion matrix, precision, recall, and F1-score.

✅ Case Study 7: Online Loan Approval System

Scenario:
A bank uses AI to predict whether a customer is eligible for a loan. The model uses income, credit history,
and employment status as input, and outputs either "Eligible" or "Not Eligible".

Q1: Is this a classification or regression problem? Explain.

Answer: Classification – because the outcome is binary/categorical.

Q2: What are the features and labels in this case?

Answer:
Features: Income, credit score, employment status, etc.
Label: Loan eligibility (Eligible / Not Eligible)

Q3: If the model begins rejecting valid applications, what steps should be taken?
Answer:

 Review and retrain with more balanced data

 Check for bias or outdated features
 Evaluate with performance metrics and improve the algorithm