An Empirical evaluation of CostBoost Extensions for Cost-Sensitive Classification

Abstract

Data Mining Technique, namely Classification, is used to predict group membership for data samples. Ensemble learning, combining multiple classifiers using bagging, boosting or stacking, are proven data-mining methods, we have used boosting in this paper for combining multiple classifiers. Cost-sensitive classification is used for classification tasks under the Cost-Based Model (CBM), unlike the Error-Based Model (EBM). EBM does not incorporate the cost of misclassifying a sample in a model building phase, while CBM does. CBM techniques usually modify the weight update equation to incorporate the misclassification cost from cost-matrix. Cost-sensitive boosters are studied and three new extensions of CostBoost algorithm CBE1, CBE2 and CBE3 are proposed and compared with existing cost based boosting classifiers. CSE1, CSE2 and CSE3 outperformed the original CostBoost by 5%, 4% and 4% respectively, in terms of misclassification cost.