A Strategy for Functional Defect Prediction in Homogenous Datasets: A case study in the SIGAA academic system

• A. Pontes UFPB
• C. Siebra UFPB
• M. Bittencourt UFPB

Abstract

The optimization of test sequences is an important resource to improve the test efficiency of complex software systems. This optimization can be carried out by means of defect prediction techniques, which are able to identify modules with a higher chance to present problems, so that these modules can be first evaluated. The current literature brings some proposals of algorithms with high accuracy for defect prediction. However they present a poor generalization power, since problems of overfitting are hidden due to the nature of the evaluation methods that are used. The aim of this work is to propose a modelling strategy based on more homogeneous datasets to trainee defect prediction models aimed at functional bugs. The object of study for evaluation of our proposal is a complex system for academic management (SIGAA), which is used in several Brazilian universities. The application in successive versions of this system shows that our proposal is able to identify the best approach for defect prediction, which in fact indicates the most problematic modules, supporting in this way the construction of optimal test sequences.

Keywords: test automation, software test, learning algorithms, Defect prediction