On Software Defect Prediction Using Machine Learning

The goal of this paper is to catalog the software defect prediction using machine learning. Over the last few years, the eld of software defect prediction has been extensively studied because of it's crucial position in the area of software reliability maintenance, software cost estimation and software quality assurance. An insurmountable problem associated with software defect prediction is the class imbalance problem. Recently, the Asymmetric Partial Least Squares Classi er (APLSC) reported by Qu et al [1] and the Kernel Principle Component Analysis (KPCA) reported in [2] have been shown to be able to tackle the class imbalance problem. Although both the APLSC and KPCA are quite novel, they have their own disadvantages: First, the APLSC is a bilinear classi er, in which the dimension is mapped to a bilinear subspace. Second, the APLSC su ers from high overlapping, especially when the data sets are nonlinear separate. The KPCA regression model does not consider the correlation between principal components and the class attribution. In this paper, we propose two kernel based classi ers, called the Asymmetric Kernel Partial Least Squares Classi er (AKPLSC) and Asymmetric Kernel Principal Component Analysis Classi er (AKPCAC) to solve the class imbalance problem. The kernel function we use is the Gaussian kernel function. Experiments are conducted on the NASA and SOFTLAB data sets. Theoretical analysis and experimental results (using F-measure, Friedman test and Tukey test) con rm the correctness and e ectiveness of our methods. As far as we know, the results presented in this paper are novel.