A Feature Selection Approach based on Grey Relational Analysis for Within-project Software Defect Prediction

摘要

Software defect prediction (SDP) is a process of predicting whether software has defects through historical data information. The existing feature selection algorithm needs to take more attention on the internal relations of features, which is important for SDP. The grey relational analysis (GRA) is introduced to describe the internal relations from the geometric similarity. The classification ability of the feature combination and the individual feature is discussed. The Banzhaf in cooperative game theory is discussed to make good evaluation on each feature. An improved feature selection approach based on One Rule (OR) algorithm for SDP is presented. The experiments are conducted on 9 datasets which are regularly used in SDP, and the results show that the proposed approach performs better than the other comparable feature selection approaches in terms of classification performance.