Recommendation of Regression Techniques for Software Maintainability Prediction With Multi-Criteria Decision-Making

Abstract

Context: Successful project management requires accurate estimation of maintenance effort and cost. Software Maintainability Prediction (SMP) plays a very important role in controlling software maintenance costs by detecting software modules with low maintainability. In previous research, numerous regression techniques were applied to predict software maintainability. The results with respect to various accuracy or performance measures are conflicting. Thus, there is a dire need to develop a method that can recommend regression techniques for predicting software maintainability in the presence of conflicting performance or accuracy measures. Objective: This paper aims to recommend suitable regression techniques for SMP based on the Multi-Criteria Decision-Making (MCDM) approach. Methodology: In our proposed approach, selecting a regression technique for SMP is modeled as the MCDM problem. To validate the proposed approach, an empirical study is done using three MCDM methods, 22 regression techniques, and eight performance measures over five software maintainability datasets. Before applying MCDM methods, a statistical test, namely the Friedman test, was conducted to ensure the significant difference between regression techniques. Results: The results of our study show that SVR, IBK, REPTree, and MLP-SVM achieve the highest-ranking score value one and are recommended as top-ranked approaches for SMP based on MCDM rankings. Conclusion: The main outcome of this study is that the proposed MCDM-based approach can be used as an efficient tool for selecting regression techniques among different available regression techniques for SMP modeling in the presence of more than one conflicting accuracy or performance measure.