Software Defect Prediction using Hybrid Approach

Defective software modules have significant impact over software quality leading to system crashes and software running error. Thus, Software Defect Prediction (SDP) mechanisms become essential part to enhance quality assurance activities, to allocate effort and resources more efficiently. Various machine learning approaches have been proposed to remove fault and unnecessary data. However, the imbalance distribution of software defects still remains as challenging task and leads to loss accuracy for most SDP methods. To overcome it, this paper proposed a hybrid method, which combine Support Vector Machine (SVM)-Radial Basis Function (RBF) as base learner for Adaptive Boost, with the use of Minimum-Redundancy-Maximum-Relevance (MRMR) feature selection. Then, the comparative analysis applied based on 5 datasets from NASA Metrics Data Program. The experimental results showed that hybrid approach with MRMR give better accuracy compared to SVM single learner, which is effective to deal with the imbalance datasets because the proposed method have good generalization and better performance measures.