Predicting maintainability of open source software using Gene Expression Programming and bad smells

Software maintenance phase of Software Development Lifecycle (SDLC) is the most expensive and complex phase that requires nearly 60-70% of the total project cost. Due to this, many software fails to get repair within real time constraint. Ascribe to technology advancements and changing requirements, software must be well developed and maintained to get adapted. Hence, it is necessary to predict software maintainability in the early phases of the lifecycle so that optimization of resources can be possible and cost can be reduced. Software Maintainability is the quality attribute of software product that explains the ease with which modifications can be performed. The main focus in this study is to propose the use of Gene Expression Programming (GEP) for the software maintainability prediction and measure its performance with various machine leaning techniques such as Decision Tree Forest, Support Vector Machine, Linear regression, Multilayer Perceptron and Radial basis function neural network. The empirical study is conducted with the help of four open source datasets. Eleven bad smells are identified and is considered as maintenance effort. Results of this study show that GEP algorithm performs better than machine learning classifiers; hence it can be used as sound alternative in the prediction of software maintainability. This study would be helpful in achieving better resource allocation hence it will be useful for developers and maintainers.