UNIT 1: INTRODUCTION TO MACHINE
LEARNING
Machine Learning (25CAH-609) | MCA with AI & ML | Chandigarh University
Course Outcome CO1: Understand basic concepts, terminology, and workflow of Machine Learning.
1. WHAT IS MACHINE LEARNING?
Machine Learning is a branch of AI that enables computers to learn from data without explicit
programming. ML systems identify patterns and improve performance through experience.
Machine Learning Overview:
2. HISTORICAL EVOLUTION & AI RELATIONSHIP
Period Key Developments Key Figures
1950s-60s Turing Test, Symbolic AI Turing, Samuel
1970s-80s Expert Systems, First AI Winter Feigenbaum
1990s-00s Statistical Learning Emphasis LeCun, Bengio, Hinton
2010s-20s Deep Learning, Big Data, Transformers Hinton, LeCun, Bengio
3. ML VS TRADITIONAL PROGRAMMING
Aspect Traditional Programming Machine Learning
Approach Rule-based, Explicit Pattern-based, Learn from data
Logic Hard-coded Discovered from data
Flexibility Fixed, Hard to modify Adapts automatically
Scalability Manual updates Improves with data
4. MACHINE LEARNING APPLICATIONS
Domain Applications
Computer Vision Face recognition, object detection, image classification
NLP Translation, sentiment analysis, chatbots, speech recognition
Recommendations Product recommendations, playlists, personalization
Healthcare Disease diagnosis, drug discovery, outcome prediction
Finance Fraud detection, credit scoring, algorithmic trading
Autonomous Self-driving cars, drones, robotics
5. KEY ML TERMINOLOGY
Term Definition
Dataset Collection of data points with features and labels
Feature Independent variable for predictions
Label/Target Dependent variable being predicted
Training Data Data to train model (70-80%)
Test Data Data to evaluate model (20-30%)
Generalization Ability to perform well on unseen data
Overfitting Model learns noise; high variance
Underfitting Model too simple; high bias
Hyperparameter Configuration set before training
6. ML PIPELINE & WORKFLOW
A typical ML project follows systematic workflow from problem definition through deployment.
6.1 Detailed Workflow Steps
Step Description Deliverable
Problem Definition Understand problem, define metrics Problem statement, KPIs
Data Collection Gather relevant data Raw dataset
Data Preprocessing Clean, handle missing values Clean dataset
Feature Engineering Select, transform features Feature matrix
Model Selection Choose algorithm(s) Candidate models
Model Training Fit model on training data Trained model
Evaluation Test on validation/test data Performance scores
Deployment Put in production Live system
7. BIAS & VARIANCE TRADEOFF
Bias-variance tradeoff is fundamental. Bias measures fit, variance measures sensitivity to data
changes.
Underfitting vs Good Fit vs Overfitting:
Learning Curves Analysis:
8. TYPES OF ML PROBLEMS
Learning Type Characteristics Examples
Supervised Learn from labeled data (X, Y) Regression, Classification
Unsupervised Find patterns in unlabeled data Clustering, Dimensionality Reduction
Semi-supervised Limited labeled + lots unlabeled Few labeled samples + many unlabeled
Reinforcement Learn through rewards/penalties Game playing, Robotics
9. ALGORITHMS OVERVIEW
Different algorithms suit different problem types. Understanding strengths is crucial for selection.
10. MODEL SELECTION & HYPERPARAMETER TUNING
Selecting right model and tuning parameters is critical for good performance.
11. LECTURES COVERED IN UNIT 1 (Lectures 1-9)
Lec Title Key Topics
1 What is ML? Definition, AI relationship, evolution
2 Motivation Data-driven decisions, why ML needed
3 Terminology Dataset, features, labels, generalization
4 Problems Regression/Classification, real examples
5 Pipeline Collection, preprocessing, evaluation
6 Methods Supervised, unsupervised, semi-supervised
7 Algorithms Clustering, reinforcement, deep learning
8 Architecture Structure, workflow steps
9 Matching Algorithms to data types, summary
12. KEY TAKEAWAYS
ML Definition: Learning from data without explicit programming. Historical Evolution: From symbolic AI
to deep learning. ML vs Programming: Flexibility and adaptability advantages. Applications: Wide range
across domains. Terminology: Features, labels, generalization critical. Pipeline: Systematic workflow
ensures success. Bias-Variance: Fundamental tradeoff to balance. Problem Types: Supervised,
unsupervised, semi-supervised, reinforcement. Algorithm Selection: Match to problem characteristics.
Continuous Improvement: Iterative feedback loops essential.
Unit 1: Introduction to Machine Learning. Foundation for MCA (AI & ML).