Using Source Code Metrics and Multivariate Adaptive Regression Splines to Predict Maintainability of Service Oriented Software
Lov Kumar, Santanu Kumar Rath, Ashish Sureka · 2017
Prediction of maintainability parameter for Object-Oriented Software using source code metrics is an area that hasattracted the attention of several researchers in academia andindustry. However, maintainability prediction of Service-Orientedsoftware is a relatively unexplored area. In this work, we conductan empirical analysis on maintainability prediction of eBay webservices using several source code metrics. We consider elevendifferent types of source code metrics as input for developinga maintainability prediction model using Multivariate AdaptiveRegression Splines (MARS) method. We compare and evaluatethe performance of the maintainability prediction model withMultivariate Linear Regression (MLR) approach and SupportVector Machine (SVM). Eight different types of feature selectiontechniques have been implemented to reduce dimension andremove irrelevant features. The experiment results reveals thatthe maintainability prediction model developed using MARSmethod achieved better performance as compared to MLR andSVM methods. Experimental results also demonstrate that themodel developed by considering a selected set of source codemetrics by feature selection technique as input achieves betterresults as compared to the approach which considers all sourcecode metrics.