EAC: Efficient Associative Classifier for Classification

Mining of large datasets which generally leads to the generation of a huge volume of rules and redundancy which Associative Classifiers find very difficult to handle. Thus, most Associative Classification algorithms like CPAR and CMAR, only mine frequent itemsets, which are then processed using additional algorithms in a greedy manner. This adds overhead in running time and makes the process more complex. In this paper, we proposed a new Associative Classification algorithm, called Efficient Associative Classifier (EAC). EAC deals with redundant association rules through an effective and simple pruning technique which also helps in cutting down the number of rules which finally form a part of the classifier. The classifiers are built in a two-phased manner so as to achieve the maximum accuracy and maximum representation of all possible class labels involved in the domain. In the first phase, association rule mining (ARM) is performed globally by taking global values of support and confidence. The rule set is pruned on the basis of information gain. Similarly, second phase deals with the ARM processed locally for each class label followed by pruning. The optimal pruned rule set is sorted based in certain parameters and supplied to the classifier. The proposed method used the information gain and entropy to build a set of optimal rules. In spite of being an efficient associative classifier, EAC achieves very good accuracy and tops the accuracy comparison charts for most of the UCI-Machine learning datasets that we used for testing.