Evidential Association Classification for High-Dimensional Data

Association classification (AC) has p-roven to be a promising approach in data mining by integrating association mining and classification. The evidential association rule-based classification (EARC), as an extension of AC in belief function framework, is an effective classification model for addressing multiple uncertainties in real-world applications, but it encounters difficulties in dealing with high-dimensional data. To adapt the EARC to high-dimensional data, an improved evidential association classification method, called EARC-HD, is developed based on four stages: entropy-based fuzzy partition, evidential class association rule mining, rule prescreening and genetic rule selection. Comparing with EARC, an entropy-based fuzzy partition strategy is designed for deriving a series of fuzzy regions, with which some irrelevant attributes can be deleted. Moreover, the number of antecedent attributes is limited for effectively reducing the computational complexity in rule generation. To improve the efficiency of the classifier, the techniques of redundancy reduction and subgroup discovery are used for prescreening the mined rule set before a genetic rule selection process. Experimental results based on real-world datasets demonstrate the effectiveness of the proposed method in dealing with high-dimensional problems.