Day 2: ML Algorithms & Practice

Tuesday, February 3, 2026

Overview

Day 2 is dedicated to machine learning algorithms for supply chain forecasting. We use an alternating lecture → hands-on pattern so you can immediately apply each concept.

Building on the ML foundations from Day 1, you’ll learn regularization techniques (LASSO, Ridge) and tree-based methods (Random Forest, Gradient Boosting). These ML models generate the demand forecasts that feed into the agent-based simulation models covered on Days 3-4.

Learning Objectives

By the end of Day 2, you will be able to:

  1. Implement regularized regression models (LASSO, Ridge, Elastic Net)
  2. Explain how LASSO performs automatic feature selection
  3. Build decision trees and understand their splitting logic
  4. Compare Random Forest vs Gradient Boosting approaches
  5. Apply cross-validation to select optimal model complexity
  6. Evaluate model performance using appropriate metrics
  7. Connect ML predictions to supply chain simulation inputs

Schedule

Time Session Type Lead
8:30-10:30 Regularization: LASSO & Ridge Lecture Sylvia
10:30-10:50 Break
10:50-12:30 Hands-on: Regularized Regression Practice Sylvia
12:30-1:30 Lunch
1:30-3:00 Trees & Ensembles Lecture Sylvia
3:00-3:20 Break
3:20-5:00 Hands-on: Demand Forecasting Practice Sylvia

Materials

Slides

Hands-on Notebooks

Available in both Python and R versions (see sidebar).

Day Structure: Alternating Pattern

Morning:   Lecture (LASSO)  →  Hands-on (LASSO)
                 │                    │
                 └──── Apply immediately ────┘

Afternoon: Lecture (Trees)  →  Hands-on (Forecasting)
                 │                    │
                 └──── Full ML pipeline ────┘

This alternating approach ensures you: - Apply each concept immediately after learning it - Build intuition through hands-on practice - Complete a full ML pipeline by end of day

Key Concepts

Morning: Regularization

Why add penalties to linear models?

Problem Solution
Many features (high-dimensional) LASSO shrinks & selects
Correlated features Ridge shrinks all
Unknown which is best Elastic Net combines both
Choosing penalty strength Cross-validation

Afternoon: Tree-Based Methods

Non-linear methods that capture complex patterns:

Method How it works
Decision Tree Recursive splitting, easy to interpret
Random Forest Many trees averaged, reduces variance
Gradient Boosting Trees that learn from mistakes

Algorithm Selection Guide

Data Characteristics Recommended Approach
Linear relationships Linear models (LASSO, Ridge)
Non-linear patterns Tree-based methods
High dimensionality (many features) LASSO, Elastic Net
Need interpretability Linear models, single decision trees
Maximum accuracy Ensemble methods (RF, Gradient Boosting)

Connection to Days 3-4

Day 2: ML Demand Forecasts  →  Day 3-4: ABM Supply Chain Simulation
    (What will demand be?)        (How will the system respond?)

Your ML predictions become inputs to the agent-based models, answering: - How much medication will each facility need? - Which facilities are at risk of stockout? - What should reorder quantities be?

Pre-Session Preparation

Before this session, ensure you:

  1. ✅ Completed Day 1 hands-on exercises (including sine wave demo)
  2. ✅ Can access Google Colab (test here)
  3. ✅ Review the glossary of ML terms

Optional Reading