Module ML10
Predictive machine learning and responsible validation
Build defensible predictive baselines before adding neural complexity.
Why this module
Build defensible predictive baselines before adding neural complexity.
What you produce
A reproducible tabular prediction study. Assess the output against the unit outcomes, explanation, safe handling, and independent transfer. Do not require public sharing or use of real sensitive records.
Chapters and units
ML10-C01: Problem and data boundaries
Choose a target and comparator that match a real decision; Fit transformations using training data only
- ML10-U01 Problem, target, and baseline Choose a target and comparator that match a real decision
- ML10-U02 Preprocessing and leakage control Fit transformations using training data only
ML10-C02: Predictive and unsupervised methods
Compare appropriate simple models; Explain when labels are absent and evaluation is difficult
- ML10-U03 Regression, classification, and trees Compare appropriate simple models
- ML10-U04 Unsupervised learning and anomaly methods Explain when labels are absent and evaluation is difficult
ML10-C03: Decision errors and final validation
Analyze decision errors across relevant slices; Protect a held-out evaluation and report limits
- ML10-U05 Imbalance, calibration, and explanation Analyze decision errors across relevant slices
- ML10-U06 Time-aware evaluation and final report Protect a held-out evaluation and report limits