MACHINE LEARNING
AGENDA
Introduction to Machine Learning
Classical versus ML programming
General Framework of ML
Types of ML
Underfitting versus Overfitting
Bias and Variance
INTRODUCTION TO
MACHINE LEARNING
4
WHAT IS MACHINE LEARNING?
• Definition: Machine Learning (ML) is a field of
study that enables computers to learn from
data without being explicitly programmed.
• Core Idea: Systems improve performance (P)
at a task (T) with experience (E)
• Why it Matters:
• Automates decision-making
• Enhances predictions
• Enables intelligent systems
CLASSICAL PROGRAMING 5
VERSUS MACHINE LEARNING
PROGRAMMING
Classical ML programming
Programming • Learns rules from
• Rigid rules data
• Manually • Requires training
hardcoded logic on datasets
• Does not adapt • Adapts with new
with data data
6
GENERAL FRAMEWORK OF ML 7
ETP Definition: ML is the study of algorithms that improve their Performance
(P) at some Task (T) with Experience (E).
Key Components:
• Task (T): The objective or problem the algorithm tries to solve
• Experience (E): Historical data used to train the model
• Performance (P): Metric to evaluate how well the model performs
Example 1: Stock Price Prediction
• T: Predict the price of a stock
• E: Historical price data
• P: Mean Squared Error (MSE)
Example 2: Customer Segmentation
• T: Segment customers into groups
• E: Transactional data
• P: Clustering metrics like intra/inter-cluster variation
8
CATEGORIES OF ML:
•By Learning Type:
•Supervised Learning: Uses labeled data (Y present)
•Unsupervised Learning: No labels (Y absent)
•Reinforcement Learning: Agent interacts with environment to
maximize reward.
•By Task Type:
•Classification: Categorize data
•Regression: Predict continuous values
•Clustering: Group similar data
•Recommendation: Suggest relevant items
•Forecasting: Predict future trends
9
10
11
UNDERFIITING V/S OVERFITTING
12
UNDERFITTING VERSUS OVERFITTING
• Underfitting:
• Fails to capture data patterns
• High training and testing loss
• Overfitting:
• Learns noise and outliers
• Very low training loss but high testing loss
• Ideal Model: Balanced tradeoff → low training and testing error
13
BIAS AND VARIANCE
•Bias: Error due to overly simplistic model
(missed patterns)
•Variance: Error due to overly complex model
(captures noise)
•Perfect Model: Low bias + low variance
•Goal: Find the sweet spot between bias and
variance
14
VISUALIZING BIAS AND VARIANCE
• Target Shooting Analogy:
• High Bias: Shots grouped but far
from bullseye → wrong aim
• High Variance: Shots scattered →
inconsistent aim
• Low Bias & Low Variance: Shots
tightly grouped around bullseye
• Underfitting = High Bias, Low
Variance
• Overfitting = Low Bias, High
Variance
15
FINAL TAKEAWAYS
•Machine Learning enables systems to learn from data and improve
performance automatically.
•Choose ML over classical programming when patterns are complex
or unclear.
•A solid ML framework relies on ETP: Experience, Task, and
Performance.
•Understand ML model categories by type of learning and task.
•Avoid underfitting and overfitting by managing model complexity.
•Grasp the bias-variance tradeoff to build models that generalize well.
•Best practices like cross-validation and regularization lead to better
performance and robustness.
THANK
YOU
Er. Palak Vyas
[email protected]+91-9826085592
