Associative Text Categorisation Rules Pruning Method

In this paper, the problem of rule pruning in associative text categorisation is investigated. We propose a new rule pruning method within an existing associative classification algorithm called MCAR. Experimental results against large text collection (Reuters-21578) using the developed pruning method as well as other known existing methods (Database coverage, lazy pruning) are conducted. The bases of the experiments are the classification accuracy and the number of generated rules. The results derived show that the proposed rule pruning method derives higher quality and more scalable classifiers than those produced by lazy and database coverage pruning approaches. In addition, the number of rules generated by the developed pruning procedure is usually less than those of lazy pruning and database coverage heuristics.