Design and implementation of improved AL based defect prediction model for industrial use

Abstract

Software systems are driving the quality of human life. Annual investment in developing newlinesoftware systems is exceeding 4 trillion USD. Given the increasing complexity of newlinesoftware products and cognitive nature of the software development process, the newlinesoftware defects are on the rise, resulting in production of poor-quality software that newlineneeds about 2.8 trillion USD every year to fix. Over the years, academic research into newlinesoftware engineering methodology has attempted improving software quality. In fact, newlineresearch into finding relation between software features and software quality dates back newlineinto 1980s. Initially, statistical techniques were helpful to find such relationships. With newlinethe re-emergence of artificial intelligence, sophisticated machine learning techniques are newlinebeing used to create models that predict presence and count of defects in a given set of newlinesoftware modules. This area of research is now known as software defect prediction newlineand abbreviated as SDP. Although not standardized, SDP follows a seven-step process newline data collection, feature-selection, model building, model training, model evaluation, newlinemodel execution and model performance evaluation. It produces about ten outputs newlinepopular of which includes binary classification of software modules (defective or nondefective), newlineestimation of number of defects in each module, and determination of defect newlineseverity. Generating such predictions early in the software development life cycle helps newlinesoftware development teams to prepare for efficient defect removal contributing to newlineincreased reliability of software under development. newline