Predicting Software Maintenance effort using neural networks

Software Maintenance is an important phase of software development lifecycle, which starts once the software has been deployed at the customer´s end. A lot of maintenance effort is required to change the software after it is in operation. Therefore, predicting the effort and cost associated with the maintenance activities such as correcting and fixing the defects has become one of the key issues that need to be analyzed for effective resource allocation and decision-making. In view of this issue, we have developed a model based on text mining techniques using machine learning method namely, Radial Basis Function of neural network. We apply text mining techniques to identify the relevant attributes from defect reports and relate these relevant attributes to software maintenance effort prediction. The proposed model is validated using `Browser´ application package of Android Operating System. Receiver Operating Characteristics (ROC) analysis is done to interpret the results obtained from model prediction by using the value of Area Under the Curve (AUC), sensitivity and a suitable threshold criterion known as the cut-off point. It is evident from the results that the performance of the model is dependent on the number of words considered for classification and therefore shows the best results with respect to top-100 words. The performance is irrespective of the type of effort category.