Gradient Tree Boosting Approach for Software Defect Prediction

Abstract

Prediction of the actual defect in software has remained a challenging task for the software developers. Any deviation in counting the no. of defects or estimating the defects may lead to serious problems like the unexpected outcome. Majorly, defects like time, cost and effort have to be computed effectively at the initial phase of software development. Some of the early developed data mining approaches are developed for quality analysis and defect prediction. But for large scale software development, these techniques are not performing well due to the high nonlinearity nature of data. This paper proposes a novel gradient boosting based machine learning approach for the effective prediction of a software defect. The proposed method has been analyzed with various performance-related factors and found to be superior among nine competitive machine learning approaches.