Does Code Complexity Affect the Quality of Real-Time Projects?
Archana Patnaik, Neelamadhab Padhy
Abstract
Code smell targets to identify bugs that occur due to incorrect analysis of code during software development life cycle. It is the task of analyzing a code design problem. The primary causes of code smell are complexity in structural design, violation of programming paradigm, and lack of unit-level testing by the software programmer. Our research focuses on the identification of code smell using different machine learning classifiers. We have considered 15 software code metrics of the Junit open source project and developed a hybrid model for code smell detection. Our dataset consists of 45 features which is further reduced by 15 using various feature selection techniques. Random sampling is used to handle the imbalance in the dataset. The project's performance is evaluated using 10 machine learning techniques which including regression, ensemble methods, and classification. Based on the statistical analysis, it is analyzed that the Random forest ensemble classifiers give best result with an accuracy of 99.12 % is the most appropriate technique for detecting different types of bad smells like god class, duplicate code, long method, large class, and refused bequest.