Directed Acyclic Program Graph Applied to Supervised Classification
Thibaut Bellanger, Matthieu Le Berre, Manuel Clergue, Jin‐Kao Hao · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2024
In the realm of Machine Learning, the pursuit of simpler yet effective models has led to increased interest in decision trees due to their interpretability and efficiency. However, their inherent simplicity often limits their ability to handle intricate patterns in data. This paper introduces a novel approach termed Directed Acyclic Graphs of Programs, inspired by evolutionary strategies, to address this challenge. By iteratively constructing program graphs from binary decision makers, our method offers a balance of simplicity and performance for classification tasks. Notably, we emphasize the preservation of model interpretability and expressiveness, avoiding the use of ensemble techniques like voting. Experimental evaluations demonstrate the superiority of our approach over existing methods in terms of both effectiveness and interpretability.