Search-Based Predictive Modelling for Software Engineering: How Far Have We Gone?

Abstract

In this keynote I introduce the use of Predictive Analytics for Software Engineering (SE) and then focus on the use of search-based heuristics to tackle long-standing SE prediction problems including (but not limited to) software development effort estimation and software defect prediction. I review recent research in Search-Based Predictive Modelling for SE in order to assess the maturity of the field and point out promising research directions. I conclude my keynote by discussing best practices for a rigorous and realistic empirical evaluation of search-based predictive models, a condicio sine qua non to facilitate the adoption of prediction models in software industry practices.