Nothing else matters: what predictive model should I use?

  • Authors:

  • Massimiliano Di Penta

  • Affiliations:

  • University of Sannio, Benevento, Italy

  • Venue:
  • Proceedings of the 7th International Conference on Predictive Models in Software Engineering
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

In past and recent years---also thanks to the availability of data from software repositories---several kinds of models have been built to characterize and, in some cases, to predict software change- and fault-proneness. While a wide variety of change- and fault-proneness models have been built so far, and although software repositories have opened the road towards promising research directions, there are several issues still to be solved. First, we need to carefully assess and validate the quality of data sets used for our models. Second, although predictive or explanatory models do not allow to claim for causation, we need to better exploit software repositories with the aim of providing qualitative, credible explanations to the statistical correlations captured by the models. Third, and most important, when building predictive models we often tend to forget what would be their ultimate usage, i.e., providing advices and recommendations to developers, with the aim of making their job easier and helping them to release more reliable software. Thus, assessing models' usability is crucial to favor their adoption.