Software Defect Prediction using KPCA & CSANFIS

Abstract

Identifying Bugs/Defects in the early stages of software life cycle reduces the effort required in software development. A lot of research has been progressed in predicting software defects using machine learning approaches. In software defect prediction, there are mainly two problems, dimensionality reduction and class Iimbalance. In this paper, we are addressing dimensionality reduction using Kernal Principle Component Analysis and Class Imbalance problem using Cost sensitive Class Imbalance Problem. Kernal Principle Component Analysis transforms non linear high dimensional data into low dimensional space.Cost Sensitive Adaptive Neuro Fuzzy Inference System assigns weights to samples based on class imbalance ratio to alleviate biasing in classification towards majority class. The performance of proposed methodology is measured using Area under ROC Curve (AuC) values. We performed experimentation on Software Defect datasets downloaded from NASA Dataset repository and observed Auc values are increased with our proposed methodology by 5-6%.