Development of code smell detection model based on graph neural network to detect long method, large class, and duplicated code

This study aims to develop and evaluate a GNN-based model for automatically detecting code smells in Python by leveraging the structural information derived from Abstract Syntax Trees (ASTs). The study employed a quantitative approach by training a Graph Neural Network (GNN) model on code represented as Abstract Syntax Trees (ASTs). The proposed GNN model demonstrated high effectiveness in detecting specific types of code smells using a processed dataset, achieving 96.3% accuracy for Long Method detection and 95.2% for Large Class detection. Furthermore, the model effectively identified Long Method, Large Class, and Duplicated Code when tested on real-world datasets. On the ERPNext dataset, it reached 95.95% accuracy, 99.84% precision, 95.95% recall, and 97.78% F1-score. On the Odoo dataset, it attained 93.29% accuracy, 99.74% precision, 93.29% recall, and 96.29% F1-score. These results show that the proposed GNN model outperformed traditional machine learning algorithms such as Decision Tree, Random Forest, Support Vector Machine (SVM), Stochastic Gradient Descent (SGD), Multi-Layer Perceptron (MLP), and Linear Regression. The results confirm that the GNN-based approach effectively detects code smells in Python programs, surpassing classical machine learning techniques. This model provides a practical tool for enhancing code quality and maintainability. Future work could explore real-time integration of this model into development environments and expand detection to other code smell types.