Predicting Code Smells: A Comparative Study of Machine Learning Classifiers for God Class and Data Class

Code Smells often indicate underlying problems that can degrade various aspects of software quality. It refers to the area in the source code which needs improvement. Software with Code Smell may lead to high maintenance costs. Code Smells are not errors or bugs; rather addressing them can enhance the code quality. Code with Smells may lead to software defects in the future. Objective- This paper aims to predict specific Code Smells, namely God Class and Data Class, present in the source code and compare the performances of various machine learning models. Methodology- We compared the performance of four traditional machine learning classifiers (Decision Tree, Naive Bayes, Logistic Regression, and K-Nearest Neighbor) with five ensemble classifiers (Random Forest, Bagging with Decision Tree, Gradient Boosting, AdaBoost, and XGBoost). Result & Conclusion- Our analysis indicates that the existence of God Class is more prevalent than Data Class in the examined code, with accuracy as the primary performance metric. Our experimental observation reveals that XGBoost outperforms predicting the Code Smells in terms of all performance measures i.e. 100%.