Prediction of Software Development Modification Effort Enhanced by a Genetic Algorithm

Gergő Balogh, Ádám Zoltán Végh and Árpád Beszédes
During the planning, development, and maintenance of software projects one of the main challenges is to accurately predict the modification cost of a particular piece of code. Several methods are traditionally applied for this purpose and many of them are based on static code investigation. We experimented with a combined use of product and process attributes (metrics) to improve cost prediction, and we applied machine learning to this end. The method depends on several important parameters which can significantly influence the success of the learning model. In the present work, we overview the usage of search based methods (one genetic algorithm in particular) to calibrate these parameters. For the first set of experiments four industrial projects were analysed, and the accuracy of the predictions was compared to previous results. We found that by calibrating the parameters using search based methods we could achieve signicant improvement in the overall efficiency of the prediction, from about 50% to 70% (F-measure).

Keywords: software development, effort estimation, modification effort, genetic algorithm.