Associative Classification using Common Instances among Conflicting Discriminative Patterns

This paper proposes a white-box, associative classifier that uses discriminative patterns mined from a dataset including numeric values. In the proposed model, when there exist conflicting patterns for a test instance, we take into account the training instances covered in common by them as the “neighbors” of the test instance. By this design, we can accurately capture the space around the test instance, and as a result, it is observed in our experiments that the predictive performance is improved from some simpler methods. We also show another advantage of the proposed classifier by inspecting its interpretability/explanability.