Kernel Based Asymmetric Learning for Software Defect Prediction

Software defect prediction is to predict the defect-prone modules for the next release of software or cross project software. Real world data mining applications, including software defect prediction domain, must address the issue of learning from imbalanced data sets. As pointed out by Khoshgoftaar et al. [1] and Menzies et al. [2], the majority of defects in a software system are located in a small percentage of the program modules, software defect data sets are highly class imbalanced. Existing approaches to solving class imbalanced problem mainly include data sampling methods and adaptive algorithm methods. The results reported in [3] show that AdaBoost almost always outperforms even the best data sampling techniques in software defect prediction. AdaBoost is a typical adaptive algorithm, and has received a good deal of attention since being introduced by Freund and Schapire [4]. It attempts to reduce the bias generated by majority class data, by updating the weights of instances dynamically according to the errors in previous learning. Besides these methods, some studies improved dimension reduction methods for the class imbalanced problem [5], [6], [7]. Most recently, Qu et al. [7] proposed an asymmetric partial least squares classifier (APLSC) to tackle the class imbalance problem. They suggested that APLSC outperform other existing algorithms, because it can extract favorable features for unbalanced classification. However it is a bilinear classifier, in which the dimension is mapped to a bilinear subspace. In this paper, we develop a kernel based asymmetric learning method, called Asymmetric Kernel Principal Component Classification (AKPCC), which is more adaptive to general situations.