Maintainability Prediction from Project Metrics Data Analysis Using Artificial Neural Network: An Interdisciplinary Study

Software maintainability is an important aspect for all software engineering paradigms. Considering the maintainability a factor influencing the software quality and reliability, the estimation can help to improve overall software quality. Maintainability is an indirect and derived measure which needs to predict using the other direct measures. Soft computing approaches have been used widely in prediction of software entities. The paper analyzes the project history data with the help of artificial neural network and produces the predicted maintainability value of the software module or component. From the project metrics data, four influencing factors identified and neural network model is built for maintainability prediction. The four simple input factors, multiple condition count, node count, percentage comments and total lines of code can be easily calculated from the source code of the module or component of a project. The less complexity of the input attributes makes the model more applicable in software industries. The ANN model is evaluated and validated on history data from three projects. The root mean square error value shows the ANN as good technique to predict the maintainability from the history data.