Adapting explainable Decision Sets to Continuous-based Real-World Problems

For machine learning (ML) models to be trusted in decision-making, eXplainable AI (XAI) becomes essential. A promising route to XAI is the development of interpretable ML models, e.g. Decision Sets (DS), which apply a swarm of rules for intelligent problem resolution. Although there has been progress in creating explainable DS using Evolutionary Computation (EC), these approaches often struggle with complex problems involving continuous features. This difficulty arises because the stochastic nature of these approaches cannot adapt to the complexity of problems based on continuous features. In response, we propose a groundbreaking method that formulates DS by breaking down the problem into sub-tasks, each addressed by a specific rule, through a novel interpretable framework. By leveraging a divide & conquer tactic and utilizing the inherent distribution of data to direct the EC search, this method assembles a rule-based intelligent model. Through testing on a wide array of datasets, our approach has proven to generate explainable DSs that maintain high prediction accuracy when dealing with complex issues involving continuous features.