A study of dealing class imbalance problem with machine learning methods for code smell severity detection using...

Area of Science:

• Software Engineering
• Machine Learning
• Data Science

Background:

• Code smells indicate design flaws, increasing maintenance costs and reducing software quality.
• High-severity code smells pose significant challenges to system maintainability, necessitating accurate severity assessment for refactoring.
• Class imbalance complicates precise code smell severity detection.

Purpose of the Study:

• To detect code smell severity using machine learning techniques.
• To address class imbalance in code smell datasets using Synthetic Minority Oversampling Technique (SMOTE).
• To evaluate the effectiveness of feature selection and machine learning models in classifying code smell severity.

Main Methods:

• Utilized four code smell severity datasets: Data class, God class, Feature envy, and Long method.
• Applied Principal Component Analysis (PCA) for feature selection and SMOTE for class balancing.
• Employed five machine learning models: K-nearest neighbor, Random forest, Decision tree, Multi-layer Perceptron, and Logistic Regression.

Main Results:

• Achieved a 0.99 accuracy score with Random Forest and Decision Tree models for the Long method code smell.
• Evaluated model performance using accuracy, Precision, Recall, and F-measure.
• Compared the impact of SMOTE on model performance, demonstrating its benefits.

Conclusions:

• The developed models show promising results for accurate code smell severity detection.
• The study highlights the importance of addressing class imbalance and employing feature selection for improved performance.
• Findings can guide future research in automated software quality assessment and refactoring prioritization.