LAUSR

Abstract With the increasing complexity of the software systems nowadays, the trend has been shifted to object-oriented (OO) development. The classes are the central construct in an OO software that are expected to be of utmost quality and high maintainability. The maintainability of a class is the probability that a class can be effortlessly modifiable in the maintenance phase. Unfortunately, it is very tough to determine the maintainability of a class with confidence before the release of the software. However, maintainability can be predicted with the help of internal quality attributes (viz. complexity, cohesion coupling, inheritance, etc.). The researchers in the literature have studied the relation amongst the internal quality attributes and class maintainability. Many class maintainability prediction models have been developed in the past with the help of internal quality attributes. Effective prediction models are vital to forecast class maintainability accurately. However, various datasets used to build prediction models for class maintainability suffer from imbalanced data problem. In that scenario, a model trained with imbalanced data gives erroneous predictions of class maintainability, which results in the inaccurate allocation of testing and maintenance resources to the misclassified classes. Therefore towards this direction, this study assesses the applicability of techniques to take care of imbalanced data. In this study the imbalanced data is treated with nine oversampling and three undersampling methods. A comprehensive comparison of fourteen machine learning (ML) techniques and fourteen search based (SB) techniques is conducted for class maintainability prediction. The results of the study support the applicability Safe-Level Synthetic Minority Oversampling Technique (Safe-SMOTE) to handle the imbalanced data for class maintainability prediction.