Code smell severity detection using machine learning

Code smells are indications of design issues in software. The importance of early detection of these indicators helps in improving the maintenance process and raising the efficiency of software work. Besides detecting the severity of code smells the big problems of refactoring can be avoided as well. Machine learning algorithms are an effective way to detect code smells. This paper explicitly examines three different algorithms for SMO, ANN, and J48, among the four, most frequently detected code smells such as data class, god class, feature envy, and along method. The results showed that the SMO algorithm got better accuracy for data class, god class, and feature envy. Apart from this, ANNE with SMO algorithms got identical higher accuracy results in the long method.