Multiple near optimal solution through seeded genetic programming for predicting software maintenance effort

In Software Engineering, a variable mostly predicted into the Software Planning is the effort (i.e., the number of person-hours) needed to develop a software project. The types of software development can be either new or maintenance. A type of software maintenance is enhancement, which corresponds to the type having the highest impact on business. An accurate prediction is needed for software managers such that a suitable budget can be estimated. Therefore, in the present study, we propose the application of a new model termed Multiple Near Optimal Solution through Seeded Genetic Programming (MUNOS–SGP) for predicting software enhancement maintenance effort. MUNOS–SGP differs from the Koza–GP, and Poli–GP algorithms in (1) how the sets are generated by execution, (2) the size of initial individuals, and (3) the number of the best individuals selected by generation during the validation method. We used nine data sets selected from an international public repository of software projects to train and test the models through a leave-one-out cross validation method. Results showed that (1) MUNOS–SGP was statistically better than a statistical regression model, Poli-GP, and Multi-layer perceptron (MLP) neural network in the nine data sets of software projects, and (2) MUNOS–SGP was statistically better than Koza-GP and support vector regression (SVR) in seven data sets, and statistically equal than Koza-GP and SVR in the remaining two data sets. Thus, we can conclude that MUNOS–SGP can be used to predict the software enhancement maintenance effort.