Information-based rule ranking for associative classification

Classification rule mining is a promising approach in data mining to create more interpretable and accurate prediction systems. This approach typically builds on top of well-known association rule mining and classification techniques, which identify a subset of rules known as class association rules (CAR), whose consequents are limited to target class labels. Existing classification rule mining methods have proven to provide better predictive accuracy while improving the interpretability and reasoning of a problem. Nevertheless, the challenges of such methods are mainly on a large number of generated CAR and the ranking and selection of interesting CAR for building classifiers. This paper proposed a hybrid of association rule mining (FP-growth) and neural network (sequential network of dense layers) techniques, focusing on using an information-based approach to rank and select interesting CAR. Preliminary experiments were conducted on nine UCI Machine Learning Repository datasets to examine the effect of the proposed hybrid model on generic datasets. The results show that the proposed approach achieved higher accuracy than other associative classification methods.