/evaluating-machine-learning-models

name: evaluating-machine-learning-models
description: |
  Evaluate trained machine learning models with the right metrics and comparison logic.
  Use for benchmark review, threshold selection, calibration, validation, and model comparison; not for feature engineering or leakage auditing.
allowed-tools: Read, Write, Edit, Grep, Glob, Bash(cmd:*)
version: 1.0.0
author: Jeremy Longshore <jeremy@intentsolutions.io>
license: MIT

Model Evaluation Suite

Use this skill when the model exists and the question is whether it is good enough.

Overview

This skill focuses on choosing and interpreting the right evaluation metrics for the problem, then comparing candidate models or thresholds.

When to Use This Skill

• Comparing candidate models with consistent metrics
• Reviewing precision/recall/F1/AUC, regression error, calibration, or ranking quality
• Stress-testing validation strategy before deployment or publication

Not For / Boundaries

• Building the training pipeline itself: use `scikit-learn` for classical modeling or `ml-pipeline-workflow` for end-to-end workflow ownership
• Engineering features: use `preprocessing-data-with-automated-pipelines`
• Checking train/test contamination: use `ml-data-leakage-guard`

Typical Outputs

• Metric suite recommendations
• Model comparison tables
• Notes on threshold tradeoffs, calibration, and validation weaknesses

Related Skills

• `scikit-learn` for class-level error breakdowns and confusion matrices
• `scientific-reporting` when the evaluation must become a deliverable