Hybrid Rule-Based Classification by Integrating Expert Knowledge and Data

The common methods for dealing with classification problems include data-driven models and knowledge-driven models. Recently, some methods were proposed to combine the data-driven model with the knowledge-driven model to construct a hybrid model, which improves the classification performance by complementing each other. However, most of the existing methods just assume that the expert knowledge is known in advance, and do not indicate how to obtain it. To this end, this paper proposes a way to obtain knowledge from experts represented by rules through active learning. Then, a hybrid rule-based classification model is developed by integrating the knowledge-driven rule base and the rule base learned from the training data using genetic algorithm. Experiments based on real datasets demonstrate the superiority of the proposed classification model.