New Associative Classification Method Based on Rule Pruning for Classification of Datasets

In data mining, a rule-based classification approach called Associative Classification (AC) normally builds accurate classifiers from supervised learning data sets. It extracts “If-Then” rules and associates each of the generated rules with two computed parameters; support and confidence. These two parameters are utilized to differentiate the rules' superiority during the building of a classifier's step. In current AC algorithms, whenever a rule is inserted into a classifier, all of its corresponding training data is discarded. However, the discarded data actually are used to compute support and confidence of other rules and will affect other lower ranked rules since rules normally have common training data examples. The use of static support and confidence may result in very large less-accurate classifiers.Thus, a procedure that amends other rules' support and confidence is important. This paper proposes a new procedure named Active Pruning Rules (APR) to overcome the above problem so then the classifiers' performance - especially predictive accuracy and reducing rule redundancy - will be further improved. The experimental results obtained from a number of University of California Irvine (UCI) data sets and real adult autism classification data set showed that APR is highly competitive to other AC and rule-based classifiers and often produces smaller yet more predictive classifiers.