An Experimental Evaluation on Feature Subset Selection using Association Rule Mining and Clusters

Feature Subset Selection (FSS) is to select a subset of features from the feature space taking into account its ability to predict future cases. In this task redundant information, irrelevant attributes, attributes that does not describe trivial information, imprecise data, and inconsistent information are removed. This work reviews several techniques found in literature and a comparison of the various algorithms is produced. A new technique is also proposed that uses Clusters and Association Rule Mining (ARM) to extract the most relevant features from different data sets. The algorithm decreases the computational cost, error rate and works with reduced time delay. Two data sets are tested using binary class problem. The proposed technique is evaluated using weka classifiers. Experimental results demonstrate that by applying this technique when employing less number of features information content of messages increase to probe significant rules