scientific-feature-importance

You can explain machine learning model predictions by quantifying individual feature contributions using tree-based importance (MDI) and permutation importance, generating visualizations and result tables for robust explanations.

MDI calculates tree-based feature importance internally, while permutation importance provides a model-agnostic assessment by validating feature impact, allowing you to compare and cross-check feature influence across datasets.

Yes, you can generate partial dependence plots and multi-target importance panels to inspect feature effects and compare feature importance contributions across multiple targets simultaneously.

You can output feature importance data, visualizations, and results tables as CSVs and figures, providing structured outputs directly suited for downstream documentation and reporting.

Yes, you need a pre-trained model on your dataset to run the provided analysis functions, which then calculate permutation importance and generate the corresponding feature importance figures and tables.