Machine learning based prediction of complex bugs in source code

Abstract

During software development and maintenance phases, the fixing of severe bugs are mostly very challenging and needs more efforts to fix them on a priority basis. Several research works have been performed using software metrics and predict fault-prone software module. In this paper, we propose an approach to categorize different types of bugs according to their severity and priority basis and then use them to label software metrics’ data. Finally, we used labeled data to train the supervised machine learning models for the prediction of fault prone software modules. Moreover, to build an effective prediction model, we used genetic algorithm to search those sets of metrics which are highly correlated with severe bugs.