Recursive Ant Colony Based ECOC: An Ensemble Learning Technique for Classifying Data

Error correcting output code (ECOC) is one of the widely used classifier ensemble technique .That technique provide solution for the various multiclass classification problem by dividing multiclass problem into binary class classification problem. In this paper, a new enhanced heuristic coding method, based on ECOC, RACS-ECOC is proposed. To generate strong classifiers for the multiclass classification problem following three steps are used. The first step starts with layered clustering-based approach. The approach can split a hard to classify multiclass problem in to subclasses and construct multiple different strong binary class classifiers on a given binary-class problems, so that the heuristic training process will not be stopped by some difficult binary-class problems. Second a weight optimization technique is used to optimise new classifiers for that problem. For that purpose a recursive ant colony algorithm is used to optimize classifiers. It optimize weights in a way that ensures the non-increasing of the heuristic training process whenever a new classifiers added to the ECOC ensemble. In third step these classifiers are updated to the classifier field to generate better classifier for the classification problem. Which provides a simple and efficient way to optimize weight and a KDD CUP intrusion detection data set is used for data mining. A performance Comparison of the proposed Recursive ant colony System-ECOC technique with existing WOLC-ECOC over the parameters like precision and recall is presented in results section.