Module 4: Data Preprocessing

Models are only as good as the data they train on. This module covers building good training datasets: handling missing values, encoding categorical data, scaling features, and selecting the ones that matter.

What You’ll Learn¶

• Identifying and imputing missing data

• Encoding ordinal and nominal categorical features

• Feature scaling: normalization vs. standardization

• L1/L2 regularization as implicit feature selection

• Sequential feature selection and feature importance with random forests

Lecture Materials¶

• 📝 Lecture Notes (PDF)

• 🧮 Math Notes (PDF)

• 📓 Chapter notebook: Building Good Training Datasets — Data Preprocessing

Lecture Videos¶

Part 1: Introduction, the Scikit-Learn API, and Missing Data | CAP 6606 — Module 4, Part 1¶

Part 2: Handling Categorical Data | CAP 6606 — Module 4, Part 2¶

Part 3: Train/Test Splits and Feature Scaling | CAP 6606 — Module 4, Part 3¶

Part 4: Selecting Meaningful Features and L1 Regularization | CAP 6606 — Module 4, Part 4¶

Part 5: Sequential Feature Selection and Feature Importance | CAP 6606 — Module 4, Part 5¶

Prerequisites¶

• Module 3

Reference¶

Raschka & Mirjalili, Machine Learning with PyTorch and Scikit-Learn, Chapter 4.