A Novel Classification Algorithm Based on Association Rules Mining

The traditional methods for mining classification rules such as heuristics or greedy methods only generate the rules that are too general or overfitting to do with the given database. Thus, they introduce high error ratio. Recently, a new method of mining classification rules is proposed: classification rules mining based on association rules (CARs). It is more advantageous than the traditional methods in that it removes noise and therefore the accuracy is higher. In this paper, we propose ECR-CARM algorithm. It is based on ECR-tree to find all CARs. Besides that, it is necessary for redundant rules pruning and rules reducing to gain the smaller rules set (i.e., reducing the time of identifying the class of new cases and increasing the accuracy). We also develop property to fast prune rules.