Simple and Accurate Classification Method Based on Class Association Rules Performs Well on Well-Known Datasets

Existing classification rule learning algorithms use mainly greedy heuristic search to find regularities in datasets for classification. In recent years, extensive research on association rule mining was performed in the machine learning community on learning rules by using exhaustive search. The main objective is to find all rules in data that satisfy the user-specified minimum support and minimum confidence constraints. Although the whole set of rules may not be used directly for accurate classification, effective and efficient classifiers have been built using these, so called, classification association rules.

In this paper, we compare “classical” classification rule learning algorithms that use greedy heuristic search to produce the final classifier with a class association rule learner that uses constrained exhaustive search to find classification rules on “well known” datasets. We propose a simple method to extract class association rules by simple pruning to form an accurate classifier. This is a preliminary study that aims to show that an adequate choice of the “right” class association rules by considering the dependent (class) attribute distribution of values can produce a compact, understandable and relatively accurate classifier. We have performed experiments on 12 datasets from UCI Machine Learning Database Repository and compared the results with well-known rule-based and tree-based classification algorithms. Experimental results show that our method was consistent and comparative with other well-known classification algorithms. Although not achieving the best results in terms of classification accuracy, our method is relatively simple and produces compact and understandable classifiers by exhaustively searching the entire example space.