XTRACTIS Features Highlight

Transparent Reverse-Engineering of an Opaque Boosted Tree Model

Benchmark vs. XTRACTIS (direct induction) and the initial Boosted Tree

Make the Reverse-engineering of a Boosted Tree model to transform this opaque model into a transparent XTRACTIS decision system that accurately makes rational and explainable predictions.

Pros & Benefits

Obtain an XTRACTIS intelligible model having comparable or better performance than the Boosted Tree (BT) model, allowing scientists and researchers to understand the internal logic of decision within the opaque BT model.

Case Hypotheses

– In this Precision Medicine application, we use the robust BT model predicting the seizure activity of a patient, selected and presented in the Use Case #28 (EEG Signal-Based Detection of Epileptic Seizures).

– For demonstration purposes, we suppose that the reference dataset that was used for the generation of this model is not available and that we only have the BT model.

TOP-MODEL INDUCTION BY REVERSE ENGINEERING OF THE BT MODEL

Intelligible Model, Explainable Decisions
  • The top-model is a decision system composed of 20 rules without chaining.
  • Each rule uses from 1 to 13 predictors among the 23 variables that XTRACTIS automatically identified as significant in the decision process (out of the 24 Potential Predictors).
  • Only a few rules are triggered at a time to compute the decision

It has a good Real Performance (on unknown data).

  • This Use Case demonstrates the ability of XTRACTIS to reveal the hidden decision logic of a BT model, while maintaining the same level of performance. This can be generalized to any type of opaque model.
  • Compared to the XTRACTIS model directly induced from the original learning dataset, intelligibility and performance levels are obviously degraded.
  • It is better to induce an XTRACTIS model from the original learning dataset if available (as in UC#28).

BT=Boosted Trees
XTRACTIS-REV of BT=XTRACTIS model induced by Reverse-engineering of the BT model

Detailed results and explanations in full document

Use Case 2025/06 (v1.0)

Powered by XTRACTIS® REVEAL v13.2.54674 (2025/02)

Download Document

CONTENTS

  1. Problem Definition
  2. XTRACTIS Reverse-engineering Process
  3. The initial Boosted Tree Model
  4. Top-Model Induction by Reverse-engineering of the BT
  5. Top-IVEs Benchmark
  6. Quantitative Metrics