Application of SMOTE and LSSVM with various kernels for predicting refactoring at method level

Abstract

Improving maintainability by refactoring is essentially being considered as one of the important aspect of software development. For large and complex systems, identification of code segments, which require re-factorization is a compelling task for software developers. Development of recommendation systems for suggesting methods, which require refactoring are achieved using this research work. Materials and Methods: Literature works considered source code metrics for object-oriented software systems in order to measure the complexity of a software. In order to predict the need of refactoring, the proposed system computes twenty-five different source code metrics at the method level and utilize them as features in a machine learning framework. An open source dataset consisting of five different software systems is being considered for conducting a series of experiments in order to assess the performance of proposed approach. LSSVM with SMOTE data imbalance technique are being utilized in order to overcome the class imbalance problem. Conclusion: Analysis of the results reveals that LS-SVM with RBF kernel using SMOTE results in the best performance.