Module 1: Interactive Class - Linear Regression

Today's Mission: Save RetailCo $10 Million

You've just been hired as a data scientist at RetailCo, a company losing $10 million annually due to poor inventory predictions. By the end of this session, you'll build a machine learning model that could reduce these losses by 40% - that's $4 million in recovered revenue. Ready to prove your worth?

Step 1: Discovering the Problem

RetailCo's CEO shows you last quarter's data: "We're either overstocking (tying up capital) or understocking (losing sales). Can machine learning help?"

30 sec

Quick Decision: What's the Real Problem Here?

Understanding why sales vary

Predicting future sales accurately

Finding statistical relationships

Testing sales hypotheses

💡 Insight

Correct! RetailCo doesn't need to understand WHY sales happen - they need to PREDICT what will happen. This is the fundamental shift from statistics to machine learning. Let's explore their data...

📦 Overstocking Case

Store #42: Ordered 10,000 units, sold 3,000

Loss: $70,000

📉 Understocking Case

Store #18: Ordered 2,000 units, could have sold 8,000

Lost Revenue: $120,000

✨ Perfect Prediction

Store #7: Ordered 5,500 units, sold 5,483

Efficiency: 99.7%

Step 2: Choosing Your Weapon

The board wants answers. You have two approaches available. Try both and see which helps RetailCo more:

🧪 Experiment: Adjust the Scenario

Data Complexity 5

Noise Level 30%

Sample Size 200

Step 3: Building Your Prediction Engine

Time to build the actual model. But remember - RetailCo needs this to work on NEW stores, not just explain past performance.

Critical Decision: How Should We Validate?

Use all data for training

Split: 80% train, 20% test

Test on training data

Use last year's data

# Build your model here
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LinearRegression

# Your code:
X = retail_data[['store_size', 'population', 'competition', 'marketing']]
y = retail_data['sales']

# Split the data (fill in the blanks)
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=___, random_state=42
)

# Train the model
model = LinearRegression()
model.fit(___, ___)

# Make predictions
predictions = model.predict(___)

🎯 Make a Prediction for a New Store

Step 4: The Overfitting Trap

Your model looks perfect on training data (R² = 0.95), but the CFO asks: "What about new stores?" Let's test it...

⚠️ Overfitting Demonstration

Training Performance

R² = 0.00

Test Performance

R² = 0.00

🛡️ Apply Regularization

Regularization prevents overfitting by penalizing complexity. Adjust the strength:

No Regularization λ = 0 Strong Regularization

Step 5: Quantifying Your Impact

Time to present to the board. Let's calculate the real business value of your model:

Your Model's Performance

92%

Prediction Accuracy

💰 Calculate Financial Impact

Enter RetailCo's parameters to see your model's value:

🏆 Mission Accomplished!

You've successfully transformed RetailCo's inventory management using machine learning. Here's what you achieved:

Key Learnings Applied

Paradigm Shift: Moved from explaining relationships to predicting outcomes
Proper Validation: Used train-test split to ensure real-world performance
Overfitting Prevention: Applied regularization for robust predictions
Business Impact: Quantified value in dollars, not just accuracy metrics

🚀 Module 1: The $10 Million Prediction Problem