An Empirical Framework for Software Aging-Related Bug Prediction using Weighted Extreme Learning Machine

Abstract

Software ageing (SA) related bugs highlight the issue of software failure within continuously running systems, resulting in a decline in quality, system crashes, resource misuse, and more.To mitigate these bugs, software companies employ various techniques, including code reviews, bug-tracking systems deployment, and thorough testing.Nevertheless, the identification of aging-related bugs remains challenging through these conventional approaches.To address this predicament, early prediction of the affected software regions due to runtime failures can be immensely valuable for software quality assurance teams.By accurately identifying the vulnerable areas, these teams can strategically allocate their limited resources during the testing and maintenance processes.This proactive approach ensures a more efficient and effective bug detection and resolution, enhancing overall software reliability and performance.This study aims to develop aging-related bug prediction models using source code metrics as input.In particular, our objective is to investigate metrics selections, data balancing, and weighted ELM to detect software runtime failure.Experimental results show that ELM with data imbalance SMOTE technique performs the best compared to weighted ELM for addressing the class imbalance problem.The weighted ELM and ELM + SMOTE can predict SA bugs, and these models can be applied to the future releases of software projects for online failure prediction well in advance.The experimental finding shows that the models trained using normal ELM with SMOTE data sampling techniques have significant performance improvement.