Niching-Assisted Genetic Programming for Finding Multiple High-Quality Classifiers

Peng Wang, Bing Xue, Jing J. Liang, Mengjie Zhang

Abstract: Explainable artificial intelligence (XAI) is a recent research focus, aiming to gain trust in machine learning models with clear insights into how the models make certain predictions. Due to its ability to evolve potentially interpretable classifiers, genetic programming (GP) is generally well-suited to XAI. However, many learning algorithms including GP usually learn a single best model. In practice, the best model in terms of training classification accuracy/error rate may not be the most appropriate one from the perspective of a domain expert due to overfitting and limited data. Multiple explicit and high-quality classifiers with the same training performance are therefore needed to increase the chances that the generated models will be considered more reasonable to experts. Therefore, this study designs a niching-assisted GP approach for classification. The results show that the proposed method can significantly increase the classification accuracy on most of the tested datasets. Further analysis shows that the designed algorithm can find different GP programs with the same classification performance, providing good interpretability for classification tasks.