Skip to main content

Review Model Results

After model selection and fitting, review the results to validate performance and understand feature contributions.

Factor Tables​

Factor tables display the GBM's exact representation:

  • Main effects: One table per feature showing its isolated impact
  • Interactions: Tables for feature combinations (if discovered)
  • Each table shows feature values/ranges and corresponding factors

For single-feature tables, a chart visualizes the factor pattern across the feature's range.

Feature Importance​

Gain-based importance ranking shows:

  • Which features contribute most to predictions
  • Relative importance as percentages
  • Only features actually used in the final model

This validates whether the model aligns with domain knowledge.

Automatic Feature Selection​

The L0-like regularization performs automatic feature selection:

  • Selected features: Included in final model
  • Excluded features: Did not improve cross-validation performance

If an expected feature is excluded, possible reasons:

  • Not helpful for improving model predictions
  • Insufficient variation in the data
  • Highly correlated with an included feature

Cross-Validation Performance​

The CV score shows out-of-sample performance during training. Use this to assess generalization and compare against baseline models.

Test Set Analysis (if provided)​

When test data was uploaded, the calibration chart shows predicted vs. actual values by quantile. Well-calibrated models show close alignment between predicted and actual lines.

Model Diagnostics​

Additional information displayed:

  • Number of trees
  • Regularization parameter values
  • Maximum interaction depth
  • Training vs. validation performance

Next Steps​