InteractionTransformer

InteractionTransformer extracts interaction features from random forest models and incorporates them into logistic regression to improve interpretability and predictive performance in biomedical predictive modeling.

Key Features:

• Hybrid Statistical-Machine Learning Framework: Integrates logistic regression with candidate interaction features derived from random forest models to combine statistical interpretability with machine-learning predictive capacity.
• Automated Feature Extraction: Automates identification and extraction of candidate interaction features from random forest models.
• Model Comparison Across Benchmarks: Compares logistic regression and random forest performance across 556 benchmark datasets to identify scenarios where interaction augmentation improves results.
• Hybrid Model Development: Incorporates random forest-derived interaction features into logistic regression to produce enhanced hybrid models.
• Enhanced Interpretation of Predictor-Outcome Associations: Provides clearer, interpretable estimates of associations between predictors and outcomes when logistic regression is augmented with extracted interaction features.

Scientific Applications:

• Predictive Modeling: Developing predictive models from complex biological data such as genomic or proteomic datasets.
• Interpretation of Complex Interactions: Elucidating intricate relationships between variables in biomedical data to support hypothesis generation and clinical decision-making.

Methodology:

Compare logistic regression with random forests across 556 benchmark datasets; use random forest models to capture complex interactions and extract those interactions as candidate features; integrate the extracted interaction features into logistic regression to create enhanced hybrid models.