A comparative study of four feature selection methods for associative classifiers

Feature Selection is a preprocessing step that has optimization effect in data mining. The feature set of a dataset generally contains redundant or irrelevant features in order to avoid the risk of incomplete description of instances and to provide utility to different purposes of the dataset. This may lead to an inefficient Classification rule mining process that bears with memory and time overhead. Recently developed Associative Classifiers like CBA, CMAR and CPAR are almost equal in accuracy and have outperformed traditional classifiers. CPAR has been found to be most consistently generating results with good average accuracy. So, it is selected to compare the suitability of four popular feature selection methods: GGA, SSGA, LVW and MIFS for classification of data. The Genetic algorithm is found to be the most suitable.