Guides ยท Science
ML Model Evaluation Basics
Evaluate models with the right metrics
This guide explains choosing evaluation metrics by problem type, using train/validation/test splits, avoiding leakage, and monitoring drift after deployment.
- ml evaluation
- metrics
- validation
- leakage
- drift
Match metrics to tasks
Use accuracy/recall/precision for classification, MAE/RMSE for regression, and AUC/PR as needed.
Split data properly
Use train/validation/test or cross-validation; keep temporal ordering for time series.
Avoid leakage
Ensure features are available at prediction time; separate users or time ranges to prevent cross-contamination.
Monitor after launch
Track live metrics, data drift, and recalibrate or retrain on a schedule.