GENESIM: genetic extraction of a single, interpretable model

TL;DR

GENESIM is a genetic algorithm-based method that converts ensembles of decision trees into a single, highly interpretable decision tree with improved predictive accuracy, bridging the gap between interpretability and performance.

Contribution

The paper introduces GENESIM, a novel genetic algorithm approach that transforms decision tree ensembles into a single, interpretable model with competitive accuracy.

Findings

GENESIM outperforms traditional decision tree induction on most datasets.

GENESIM achieves accuracy comparable to ensemble methods.

The resulting model is highly interpretable due to its low complexity.

Abstract

Models obtained by decision tree induction techniques excel in being interpretable.However, they can be prone to overfitting, which results in a low predictive performance. Ensemble techniques are able to achieve a higher accuracy. However, this comes at a cost of losing interpretability of the resulting model. This makes ensemble techniques impractical in applications where decision support, instead of decision making, is crucial. To bridge this gap, we present the GENESIM algorithm that transforms an ensemble of decision trees to a single decision tree with an enhanced predictive performance by using a genetic algorithm. We compared GENESIM to prevalent decision tree induction and ensemble techniques using twelve publicly available data sets. The results show that GENESIM achieves a better predictive performance on most of these data sets than decision tree induction techniques and a predictive performance in the same order of magnitude as the ensemble techniques. Moreover, the resulting model of GENESIM has a very low complexity, making it very interpretable, in contrast to ensemble techniques.