Swarm Intelligence and Variable Precision Rough Set Model: A Hybrid Approach for Classification

Abstract

Decision Tree is the most influential machine learning technique for the classification of real world datasets. Entropy based traditional Decision Tree classification methods often produce poor results due to the vagueness, uncertainty and noise in the data. In the past few decades, Rough Set Theory (RST) has been succeeded in dealing with uncertainty and vagueness. Thus, incorporating Rough Set concepts in the construction of Decision Tree for the classification of vague and uncertain data, yields better results. But RST based Decision trees sometimes classifies data too excessively and leads to a major problem called model Overfitting. This problem can be rectified by the Variable Precision Rough Set Theory (VPRST); which allows some level of uncertainty and misclassification in the construction of decision tree and classifies data more accurately. Even though the classification technique is very efficient, due to the increasing dimensionalities of the data sets the accuracy of the classification model is affected. So, from this high dimensional datasets with huge set of features, there is a need to select the most promising ones that contribute maximum for classification. In this paper, the most popular swarm intelligence technique namely Artificial Bee Colony algorithm is used for selecting the robust features and the reduced data set is submitted to the VPRST based Decision Tree for classification. In turn, a Reduced VPRSDT was obtained with promising results and it outperformed the traditional methods of tree induction, both in terms of reduced tree size and significant increase in accuracy.