AI vs ML vs DL; Machine Learning — Supervised Learning — Artificial Intelligence

AI vs ML vs DL

Hierarchy Diagram

+--------------------------------------------------------------------------+
|                    ARTIFICIAL INTELLIGENCE (AI)                          |
|         (Broad concept: making machines intelligent)                     |
|                                                                          |
|   +---------------------------------------------------------------+      |
|   |                  MACHINE LEARNING (ML)                        |      |
|   |       (Subset of AI: machines learn from data)                |      |
|   |                                                               |      |
|   |   +-------------------------------------------------------+   |      |
|   |   |               DEEP LEARNING (DL)                     |   |      |
|   |   |  (Subset of ML: uses multi-layer neural networks)    |   |      |
|   |   +-------------------------------------------------------+   |      |
|   +---------------------------------------------------------------+      |
+--------------------------------------------------------------------------+

Side-by-Side Comparison

Feature	AI	ML	DL
Definition	Machines mimicking human behavior	Machines learn from data	Multi-layer neural networks learn features
Techniques	Rules, logic, search, ML, DL	Statistical models, decision trees	CNNs, RNNs, Transformers
Data need	Varies	High	Very high
Human intervention	High (traditional AI)	Moderate	Low
Performance	Varies	Good	Best for complex tasks
Hardware	Standard CPU	GPU helpful	GPU/TPU essential
Example	Siri (overall system)	Spam filter	Facial recognition

Key Relationships

All Deep Learning = Machine Learning — DL is a type of ML
All Machine Learning = AI — ML is a way to achieve AI
Not all AI = ML — AI includes rule-based systems too
Not all ML = DL — ML also includes classical algorithms (decision trees, SVM)

Machine Learning — Supervised Learning

What is Machine Learning?

Machine Learning (ML) is a field of study that gives computers the ability to learn without being explicitly programmed (Arthur Samuel, 1959).

It involves training algorithms to recognize patterns in data and make predictions or decisions.

Mathematical foundation:

Input data: X = {x1, x2, x3, ..., xn}
Output:     Y = f(X)       (the mapping function)
Goal:       Learn function f such that predicted Y matches actual Y

What is Supervised Learning?

Supervised Learning is where you have:

Input variables (X) — known features/attributes
Output variable (Y) — known label or target value

And you use an algorithm to learn the mapping function:

Y = f(X)    (the Model / Formula)

Why "Supervised"? The algorithm learns from a labeled training dataset — like a teacher who gives correct answers while the student learns. The correct answers (labels Y) are provided and the algorithm iteratively corrects itself until it achieves acceptable accuracy.

Training Process:

Training Data (X, Y)
       ↓
Algorithm learns f(X)
       ↓
Trained Model
       ↓
Predict Y for new unseen X

Key Properties:

Output variable is already known for each input variable

Algorithm learns to map input and output

Model learns to associate features with predefined categories
Learning stops when acceptable accuracy is achieved

10.1 Regression — Predicting Continuous Values

Regression is for predicting continuous/numeric output values (future values).

Population growth prediction
Life expectancy prediction
Market forecasting/prediction
Advertising popularity prediction
Stock market price prediction

Additional Examples:

House price prediction (given area, rooms, location)

Temperature prediction

Sales revenue prediction
Car fuel efficiency prediction

Regression Algorithms:

Algorithm	Description	Use Case
Linear Regression	Fits: y = mx + b	Single feature, linear relationship
Multiple Linear Regression	y = m1x1 + m2x2 + ... + b	Multiple features
Polynomial Regression	Fits curved line: y = ax² + bx + c	Non-linear relationships
Logistic Regression	Predicts probability (0-1)	Binary classification (despite the name)
Ridge Regression	Linear + L2 regularization penalty	Prevent overfitting
Lasso Regression	Linear + L1 regularization penalty	Feature selection
Naive Bayes	Probabilistic, uses Bayes theorem	Text classification, fast
Support Vector Machine (SVM)	Finds optimal hyperplane	High-dimensional data

Linear Regression Example:

House Price = 5000 x (Area sq ft) + 200000 x (Location Factor) + base_price

If Area = 1500 sq ft, Location Factor = 1.2:
Price = 5000 x 1500 + 200000 x 1.2 = 75,00,000 + 2,40,000 = Rs. 77,40,000

10.2 Classification — Predicting Categories

Classification is for predicting which category/class an input belongs to (discrete output).

Find whether an email received is spam or ham
Identify customer segments
Find if a bank loan is granted
Identify if a kid will pass or fail in an examination

Additional Examples:

Is a tumor malignant or benign?
Is a transaction fraudulent or legitimate?
What digit is in this handwritten image (0-9)?
Which species does this plant belong to?

Classification Algorithms:

Algorithm	Description	Strength
Logistic Regression	Uses sigmoid function; outputs probability	Simple, interpretable
Decision Tree	Tree of if-else rules	Interpretable, visual
Random Forest	Ensemble of decision trees	High accuracy, robust
Support Vector Machine (SVM)	Finds hyperplane with maximum margin	High-dimensional data
K-Nearest Neighbor (KNN)	Classifies by majority vote of k nearest	Simple, non-parametric
Naive Bayes	Probabilistic classification	Fast, good for text

Binary vs Multi-class Classification:

Binary:      Spam/Ham, Yes/No, Pass/Fail, Fraud/Genuine
Multi-class: Dog/Cat/Bird/Fish, Digit (0-9), Disease type

Evaluation Metrics

Metric	Formula	When to Use
Accuracy	Correct / Total	General balanced datasets
Precision	TP / (TP + FP)	When false positives are costly
Recall	TP / (TP + FN)	When false negatives are costly
F1-Score	2 x (P x R) / (P + R)	Imbalanced datasets
MSE	Mean squared error	Regression
R² Score	Goodness of fit	Regression

AI vs ML vs DL; Machine Learning — Supervised Learning