A compact and understandable associative classifier based on overall coverage

Associative classification is a machine learning approach that aims to build accurate, effective and compact classification models (classifiers) by combining paradigms from classification and association rule mining. Research studies show that associative classification approaches could achieve higher accuracy than some of the traditional classification methods. In this paper, we propose a simple and accurate classification method by selecting “strong” class association rules that highly contribute to improve the overall coverage of the classifier. The advantage of our proposed classifier is that it generates reasonably less rules on bigger datasets compared to traditional rule-based classifiers. We also discuss how the overall coverage of such classifiers affects their classification accuracy. We have performed experiments on 15 real-life datasets from the UCI Machine Learning Database Repository and compared our proposed associative classifier with other 8 well-known classification algorithms on accuracy and the number of classification rules (all differences were tested for statistical significance). Experimental results show that our proposed method was comparative with other well-known classification algorithms on accuracy, it achieved the fourth-highest average accuracy (82.7%) among all classification methods, and tends to outperform the other algorithms in terms of average number of rules (especially on bigger datasets). Although not achieving the best results in terms of classification accuracy, our approach is relatively simple and produces a compact and understandable classifier by exhaustively searching the entire example space.

Перевод пока недоступен