Software Defect Prediction Using Automatic Feature Extraction

Abstract

Developing defect free, reliable software is a tedious task in software development process. Growing size and complexity of software makes it hard to identify faults in software modules. Early detection of defects in the development process saves significant amount of time, cost and effort. Employing fault prediction models built using machine learning helps software developers to identify errors quickly and take corrective measures. However, developing such models is difficult task since capturing the context information of source code plays a vital role in accurate prediction of bugs as demonstrated by recent studies. Automatic extraction of prediction parameters using deep learning approaches is gaining attention in this direction. This paper provides a brief survey of software defect prediction approaches, discusses the progress and current challenges in this field.