Building the hierarchical Choquet integral as an explainable AI classifier via neuroevolution and pruning

Explainability is considered essential in enabling artificial intelligence (AI) in some crucial industries, e.g., healthcare and banking. However, conventional algorithms suffer a trade-off between readability and performance, encouraging the emergence of explainable AI. In this paper, we propose a novel method to form the hierarchical Choquet integral (HCI) as an explainable AI to retain the model's accuracy and explainability. To achieve this purpose, we first adopted neuroevolution, which combines a genetic algorithm (GA) and a neural network (NN), and pruned weights to obtain information about the hierarchical decomposition of the Choquet integral. We then fine-tuned the weights of the HCI model for the classification problem. In addition, we use four datasets to illustrate the proposed algorithm and compare the results with the conventional classifiers: decision tree, deep learning, and support vector machine (SVM). The empirical results indicate that the proposed algorithm outperforms others in terms of accuracy, and keeps the Choquet integral's explainable property, justifying this paper's contribution.