Abstract
Software Defect Prediction (SDP) is a dynamic research field in the software industry. A quality software product results in customer satisfaction. However, the higher the number of user requirements, the more complex will be the software, with a correspondingly higher probability of failure. SDP is a challenging task requiring smart algorithms that can estimate the quality of a software component before it is handed over to the end-user. In this paper, we propose a hybrid approach to address this particular issue. Our approach combines the feature selection capability of the Optimized Artificial Immune Networks (Opt-aiNet) algorithm with benchmark machine-learning classifiers for the better detection of bugs in software modules. Our proposed methodology was tested and validated using 5 open-source National Aeronautics and Space Administration (NASA) data sets from the PROMISE repository: CM1, KC2, JM1, KC1 and PC1. Results were reported in terms of accuracy level and of an AUC with highest accuracy, namely, 94.82%. The results of our experiments indicate that the detection capability of benchmark classifiers can be improved by incorporating Opt-aiNet as a feature selection (FS) method.