A Hybrid Multi-Filter Wrapper Feature Selection Method for Software Defect Predictors

Abstract

Software Defect Prediction (SDP) is an approach used for identifying defect-prone software modules or components. It helps software engineer to optimally, allocate limited resources to defective software modules or components in the testing or maintenance phases of software development life cycle (SDLC). Nonetheless, the predictive performance of SDP models reckons largely on the quality of dataset utilized for training the predictive models. The high dimensionality of software metric features has been noted as a data quality problem which negatively affects the predictive performance of SDP models. Feature Selection (FS) is a well-known method for solving high dimensionality problem and can be divided into filter-based and wrapper-based methods. Filter-based FS has low computational cost, but the predictive performance of its classification algorithm on the filtered data cannot be guaranteed. On the contrary, wrapper-based FS have good predictive performance but with high computational cost and lack of generalizability. Therefore, this study proposes a hybrid multi-filter wrapper method for feature selection of relevant and irredundant features in software defect prediction. The proposed hybrid feature selection will be developed to take advantage of filter-filter and filter-wrapper relationships to give optimal feature subsets, reduce its evaluation cycle and subsequently improve SDP models overall predictive performance in terms of Accuracy, Precision and Recall values.