Support vector regression for predicting software enhancement effort

Context: Software maintenance (SM) has to be planned, which involves SM effort prediction. One type of SM is enhancement, where new functionality is added or existing functionality changed or deleted. Objective: Analyze the prediction accuracy of two types of support vector regression (ε-SVR and ʋ-SVR) when applied to predict software enhancement effort. Method: Both types of support vector regression used linear, polynomial, radial basis function, and sigmoid kernels. Prediction accuracies for ε-SVR and ʋ-SVR were compared with those of statistical regressions, neural networks, association rules, and decision trees. The models were trained and tested with five data sets of enhancement projects from Release 11 of the International Software Benchmarking Standards Group (ISBSG). Each data set was selected on the basis of data quality, development platform, programming language generation, and levels of effort recording. Results: The polynomial kernel ε-SVR (PKε-SVR) was statistically better than statistical regression, neural networks, association rules and decision trees, with 95% confidence. Conclusions: A PKε-SVR could be used for predicting software enhancement effort in mainframe platforms and coded in a third-generation programming languages, and when enhancement effort recording includes the efforts of the development team, its support personnel, the computer operations involvement, and end users.