Application of Random Vector Functional Link Network for Software Defect Prediction

Abstract

Software defect prediction (SDP) aims to develop predictive techniques that identify software modules’ default proneness using structural and quality attributes. The use of randomized neural networks, especially random vector functional link (RVFL) network, has been limited in SDP. This study proposes a novel SDP model based on principal component analysis (PCA) and RVFL. The model uses PCA for dimensionality reduction of the data and employs RVFL to classify the defective modules. Extensive experiments are conducted on 17 PROMISE repository datasets, and the proposed model is compared with ten classic machine learning (ML) techniques for within-project and inter-release scenarios. The experimental results indicate that PCA–RVFL outperforms the classic ML techniques for most of the datasets, proving the predictive capability of the proposed model in the field of SDP.