Three-way decisions based software defect prediction

Abstract

Based on a two-stage classification method and a two-stage ranking method on three-way decisions, this paper introduces a three-way decisions framework for cost-sensitive software defect prediction. For the classification problem in software defect prediction, traditional two-way decisions methods usually generate a higher classification error and more decision cost. Here, a two-stage classification method that integrates three-way decisions and ensemble learning to predict software defect is proposed. Experimental results on NASA data sets show that our method can obtain a higher accuracy and a lower decision cost. For the ranking problem in software defect prediction, a two-stage ranking method is introduced. In the first stage, all software modules are classified into three different regions based on three-way decisions. A dominance relation rough set based ranking algorithm is next applied to rank the modules in each region. Comparison experiments with 6 other ranking methods present that our proposed method can obtain a better result on FPA measure.