Defect prediction and software risk

Abstract

Defect prediction has always fascinated researchers and practitioners. The promise of being able to predict the future and act to improve it is hard to resist. However, the operational data used in predictions are treacherous and the prediction is usually done outside the context of the actual development project, making it impossible to employ it for software quality measurement or improvement. Contextualizing, imputing missing observations, and correcting operational data related to defects is essential to gauge software quality. Such augmented data can then be used with domain- and project-specific considerations to assess risk posed by code, organization, or activities and to suggest risk-specific remediation activities.