Software developers and managers struggle with the increasing number of software problems and defects. Such problems can rapidly increase the costs of software maintenance and development. Traditional defect management models cannot detect defects early enough to help developers and managers optimize costs and efforts. This is where software fault prediction comes into play. Software fault prediction “aims to identify fault-prone software modules ... before the actual testing process begins.”

This paper is a comprehensive literature review of software fault prediction techniques. It considers papers from 1993 to today. After introductory sections on software fault prediction, the paper presents three core sections on software fault datasets, methods to build software fault prediction models, and performance evaluation measures. Many datasets are described, including the features used to train predictive models and evaluation methods.

Furthermore, three sections summarize the papers included in the review. Highlights include section 8, which discusses challenges and future directions for software fault prediction. For instance, the paper argues that practitioners need more than just faulty or non-faulty information about a software component. Thus, a new brand of studies is warranted “to make fault prediction models more informative.”

The paper is well organized and comprehensive, although it is not clear how the papers included in the review were selected. Even so, any researcher working on software fault prediction will surely appreciate reading it.