Extending Decision Predicate Graphs for Comprehensive Explanation of Isolation Forest

Matteo Ceschin, Leonardo Arrighi, Luca Longo, Sylvio Barbon Junior·May 06, 2025

Summary

A novel Explainable AI method for Isolation Forest, using Decision Predicate Graphs, enhances global explainability. This approach offers insights into outlier detection, providing a comprehensive view of the decision-making process and feature usage. It promotes a fully explainable machine learning pipeline, addressing transparency, reliability, and regulatory requirements. The method contrasts Shapley Additive ExPlanations (SHAP) and Depth-based Isolation Forest Feature Importance (DIFFI), proposing a global explanation based on Decision Predicate Graphs (DPG) with an Inlier-Outlier Propagation Score (IOP-Score) metric. The text discusses applications in predictive maintenance, model interpretation, and anomaly detection, referencing works by Ndao et al., Lundberg & Lee, and Carletti et al.

Advanced features