model-evaluation
How to evaluate ML models honestly — task-appropriate metrics (classification vs regression), train/val/test splits and cross-validation, baselines, confusion matrices, class imbalance, data leakage, and overfitting vs underfitting. Use when measuring, comparing, or reporting model quality.
Included with Vanara Pro ($10/mo) — unlocks all 206 items. See pricing →
Overview
An impressive number on the wrong test set means nothing. The job of evaluation is to produce an honest, decision-grade estimate of how a model will behave on data it has never seen — and to make that estimate hard to fool, including by yourself. This skill is the deep reference: how to split data, which metric matches which task, how to read a confusion matrix, and the failure modes (leakage, imb
What it covers
- Splitting data: train / validation / test
- Pick the metric for the task
- Read the confusion matrix
- Always beat a baseline
- Class imbalance
- The Confusion Matrix, Worked
- Splitting & Cross-Validation (and Avoiding Leakage)
- Metrics by Task