iForest: Interpreting Random Forests via Visual Analytics.

Area of Science:

• Machine Learning
• Data Visualization
• Artificial Intelligence

Background:

• Random forests are powerful ensemble models comprising multiple decision trees.
• Their ensemble nature enhances predictive accuracy but compromises model interpretability.
• Lack of transparency limits applications in sensitive fields like medical diagnosis and fraud detection.

Purpose of the Study:

• To develop a visual analytic system for interpreting random forest models and their predictions.
• To address the challenge of understanding complex decision paths within random forests.
• To reduce the cognitive load on users attempting to interpret model behavior.

Main Methods:

• Proposed a visual analytic system designed for random forest interpretation.
• The system provides access to individual tree structures and properties.
• Summarizes and visualizes decision paths across the ensemble.

Main Results:

• The system effectively summarizes complex decision paths within random forests.
• Demonstrated utility through two usage scenarios.
• User study confirmed the system's effectiveness in aiding interpretation.

Conclusions:

• The developed visual analytic system enhances the interpretability of random forest models.
• Summarizing decision paths offers insights into model mechanisms.
• The system reduces the burden of interpretation for users in critical applications.