| Abstract |
To preserve software quality and maintainability, machine learning-based code smell detection has been proposed, and the results are promising. This research proposes an enhanced version of machine learning-based code smell detection. We improve the performance of machine learning-based code smell classifiers by applying hyper-parameter optimization techniques in Particle swarm optimization and Bayesian optimization to decision tree and random forest. The models were trained and evaluated on 74 open source projects to identify god class, data class, feature envy, and long method. The experimental results confirm that the optimized machine learning classifiers c an achieve up to 99.183% and 99.155% of accuracy for both class-level and function-level code smell classification, respectively. In term of recall, the enhanced machine learning-based code smell classifiers achieved 9 9.514% when identifying data class and 98.806% for long method. The comparison results also indicated that the enhanced machine learning classifiers outperform the original versions in the code smell detection context. |
Publication
Journal Publication
ไทย