Unsupervised Learning
Through Building

AI Trade School - Session 2B

Finding Patterns Without Labels

🔄 Welcome Back

Yesterday You Learned Supervised Learning

Churn Prediction

Classification Model

Price Prediction

Regression Model

Today: You'll Find Patterns Without Labels

📊 Supervised vs Unsupervised

Yesterday: We Had Labels

Churn: Yes/No | Price: $15,000

Today: No Labels

Machine Finds Patterns on Its Own

🎯 Why This Matters

80% of Business Data is UNLABELED

10,000 Customers

No pre-assigned segments

1M Transactions

No pre-marked fraud

Employee Data

No pre-identified risk

Unsupervised Learning

Finds patterns automatically

🔍 Clustering: Customer Segmentation

Problem: How many types of customers do you have?

Solution:

Feed the machine 2,000 customer records (age, income, purchase frequency, order value).

Machine discovers 4 natural customer groups.

You name them and create targeted marketing for each.

Result: Generic campaigns (2% conversion) → Targeted campaigns (6-8% conversion)

📍 The 4 Customer Clusters

Machine Discovered These Naturally

Budget Shoppers

Young (22), $35K income, 1-2 purchases/month, $25/order

Premium Buyers

Age 42, $150K income, 1-2 weekly, $200/order

Window Shoppers

Age 28, $55K income, 2-3/year, $50/order

Bulk Buyers

Age 45, $80K income, quarterly, $400/order

🏢 Real Companies Doing This

These successes are built on clustering

Amazon

"Frequently bought together"

Spotify

"Discover Weekly"

Netflix

"Because you watched X"

Every Bank

Fraud detection systems

🚨 Anomaly Detection: Finding Odd Things

Finding patterns that DON'T fit

Your Credit Card Pattern:

• $5 coffee at 8am

• $50 gas at 6pm

• $30 groceries at 7pm

Then: $500 jewelry purchase at 3am in another country

🚩 ANOMALY - Your bank freezes the card

🔍 Where Anomaly Detection Happens

Credit Card Fraud

Unusual spending patterns

Network Security

Weird login patterns = breach alert

Manufacturing

Defective products (weight, temp)

Healthcare

Unusual patient symptoms

⚙️ How Anomaly Detection Works

1

Learn Normal

Analyze thousands of regular transactions

2

Measure Distance

How different is each transaction?

3

Score & Flag

High distance = suspicious

Phase 2: BUILD

Create 2 Real ML Models

Exercise 1

Customer Segmentation (50 min)

Exercise 2

Anomaly Detection (40 min)

🎯 Exercise 1: Customer Segmentation

Business Problem:

Online retailer with 2,000 customers. Marketing needs targeted campaigns.

Question: What types of customers do we have?

Your Job: Use K-Means clustering to discover 4 customer groups

Deliverable: Name each cluster and create marketing strategy for each

Time: 50 minutes

🎯 Exercise 2: Anomaly Detection

Business Problem:

Bank processes 5,000 transactions daily. Need to find fraudulent ones.

Question: Which transactions are suspicious?

Your Job: Build anomaly detection model to flag suspicious transactions

Deliverable: List top 10 anomalies and explain why they're suspicious

Time: 40 minutes

🔧 How K-Means Works (Simple Version)

1

Tell Machine

"Find 4 groups" (K=4)

2

Place Centers

Machine puts 4 random center points

3

Assign Customers

Each to nearest center

4

Move Centers

To average of assigned customers

Repeat until centers stop moving = 4 distinct clusters!

🤔 How Many Clusters?

2 Clusters

Too broad - no useful segments

4-6 Clusters

Sweet spot - practical and useful

10+ Clusters

Too specific - hard to manage

The Answer

Platform shows you a graph to help decide

Phase 3: SHOWCASE

Present Your Models

30 Minutes

4-5 Students Present

🏆 What You Built Today

Two Production-Quality ML Models

Clustering Model

Discovered 4 customer segments

Anomaly Detection

Identified suspicious transactions

Same techniques Netflix, Amazon, and every bank use

Two Sessions. Four Models.

Session 2A: Supervised Learning

Churn Prediction | Price Prediction

Session 2B: Unsupervised Learning

Customer Segmentation | Fraud Detection

That's what real data scientists build

💡 Why This Matters

Supervised Learning

Predict known outcomes (with labels)

Unsupervised Learning

Discover hidden patterns (no labels)

80% of Data

Is unlabeled - unsupervised is MORE valuable

Your Skill

$100K-$200K starting salary skill

📚 Next Week: Deep Learning

Neural Networks

Mimicking How Human Brains Learn

Complete Module 2.2 Before Next Session

You Built

Machine Learning Models

Production-quality, real-world tools

That's not beginner work. That's pro work.

🎯 The Real Difference

When Do You Use Each?

Supervised

You have historical outcomes and want to predict future ones

Unsupervised

You want to discover hidden patterns or group similar items

Mastering both = job-ready skill

Here's The Challenge

Most people can't explain

What You Just Learned

You can. To anyone.

That confidence matters

You Completed

Session 2A + 2B

Supervised + Unsupervised Learning

You understand machine learning from BOTH sides

Thank You

For showing up and doing the work

See you next week for Deep Learning

Questions?

Unsupervised LearningThrough Building

AI Trade School - Session 2B

🔄 Welcome Back

Yesterday You Learned Supervised Learning

Churn Prediction

Price Prediction

Today: You'll Find Patterns Without Labels

📊 Supervised vs Unsupervised

Yesterday: We Had Labels

Churn: Yes/No | Price: $15,000

Today: No Labels

Machine Finds Patterns on Its Own

🎯 Why This Matters

80% of Business Data is UNLABELED

10,000 Customers

1M Transactions

Employee Data

Unsupervised Learning

🔍 Clustering: Customer Segmentation

Problem: How many types of customers do you have?

Solution:

📍 The 4 Customer Clusters

Machine Discovered These Naturally

Budget Shoppers

Premium Buyers

Window Shoppers

Bulk Buyers

🏢 Real Companies Doing This

These successes are built on clustering

Amazon

Spotify

Netflix

Every Bank

🚨 Anomaly Detection: Finding Odd Things

Finding patterns that DON'T fit

Your Credit Card Pattern:

🔍 Where Anomaly Detection Happens

Credit Card Fraud

Network Security

Manufacturing

Healthcare

⚙️ How Anomaly Detection Works

Phase 2: BUILD

Create 2 Real ML Models

Exercise 1

Exercise 2

🎯 Exercise 1: Customer Segmentation

Business Problem:

🎯 Exercise 2: Anomaly Detection

Business Problem:

🔧 How K-Means Works (Simple Version)

Repeat until centers stop moving = 4 distinct clusters!

🤔 How Many Clusters?

2 Clusters

4-6 Clusters

10+ Clusters

The Answer

Phase 3: SHOWCASE

Present Your Models

30 Minutes

🏆 What You Built Today

Two Production-Quality ML Models

Clustering Model

Anomaly Detection

Same techniques Netflix, Amazon, and every bank use

Two Sessions. Four Models.

Session 2A: Supervised Learning

Session 2B: Unsupervised Learning

That's what real data scientists build

💡 Why This Matters

Supervised Learning

Unsupervised Learning

80% of Data

Your Skill

📚 Next Week: Deep Learning

Neural Networks

Mimicking How Human Brains Learn

You Built

Machine Learning Models

Production-quality, real-world tools

Unsupervised Learning
Through Building