Software Defect Data Collection Framework for Github

Abstract

Software has become part of every sphere of life. This increasing dependence on software has put tremendous pressure on software development teams to deliver software applications as early as possible at the cost of compromised software quality and reliability. Software quality requires extensive testing and validation of software, which is not possible with limited human resources, time and budget, so researchers moved to a new paradigm of software quality assurance i.e., Software Defect Prediction (SDP). SDP aims to build automated Machine Learning (ML) models to aid development teams in prioritizing the key aspects of software testing while maintaining the short software development life cycle. SDP requires huge amount of data to train and test ML models, traditionally PROMISE and NASA defect datasets are most prominently used by researchers, but with changes in programming languages, programming styles and limited size of datasets has made them infeasible for SDP in current scenarios. In this paper, we have developed a software defect dataset collection framework, which mines commit level defect data from GitHub. The efficiency of data mining, accuracy of data and validity of data is verified by SDP models. Results shows that proposed method is feasible as well as efficient to execute even on regular computer systems.