A Survey of Machine Learning Algorithms for Defective Steel Plates Classification

This paper aims to explore various classification models and data processing techniques for classifying defects on steel plates. The study is based on a dataset of steel plates with seven different types of defects and seeks to identify the most effective combination of machine learning models and data processing techniques for automatic defect recognition. Key methodologies include data augmentation, feature selection, and dimensionality reduction techniques like SMOTE, PCA, and REF. Various models, including Naive Bayes, Logistic Regression, SVM, Decision Trees, Random Forests, and Neural Networks, are evaluated. The results indicate that Hyperparameter Tuning and REF significantly enhance model performance, with Random Forest and Neural Networks achieving the highest accuracy. Random Forest, due to its interpretability and efficiency, is identified as the best model for this classification task. This research lays a foundation for applying machine learning algorithms in industrial defect detection, promising improvements in automation, efficiency, and production quality.