Enhancing software defect prediction using supervised-learning based framework

Abstract

Software Defect Prediction (SDP) proposes to define the exposure of software to defect by building prediction models through using defect data and the software metrics with several learning algorithms which aid in identifying potentially faulty program modules, thus leading to optimal resource allocation and utilization. However, the quality of data and robustness of classifiers affect the accuracy of prediction for these models of classification compromised by data quality such as high dimensionality, class imbalance and the presence of noise in the software defect datasets. This paper presents a combined framework to enhance SDP models in which we use ranker Feature Selection (FS) techniques, Data Sampling (DS) and Iterative-Partition Filter (IPF) to defeat high dimensionality, class imbalance and noisy, respectively. The experimental results confirm that the proposed framework is effective for SDP.