Scholarlink

<p>Accurate defect detection in steel plates ensures safety, prevents damage, and reduces repair costs. This study explores ensemble machine learning techniques for defect detection, addressing the challenges of multivariate, imbalanced data with relevant and irrelevant features. Six ensemble models are evaluated on full and reduced feature sets (using Principal Component Analysis (PCA)). To mitigate data imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is applied. A publicly available dataset with 1941 instances, 34 columns, and 27 independent variables is used, comprising seven fault types. The data was split into 80% for training and 20% for testing. Results show that SMOTE significantly enhances classification accuracy, with Stacked Ensemble and XGBoost achieving near-perfect accuracies of 0.9916 and 0.9915, respectively. Additionally, SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) are used to interpret model predictions. These findings underscore the effectiveness of combining SMOTE with ensemble models in improving automated defect detection systems for steel plates.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Defect Detection in Steel Plate Manufacturing with Explainable Machine Learning and SMOTE for Imbalanced Data

  • 期刊论文

  • 发表日期：

    2025年4月10日

查看全文
  • Abdelhakim Dorbane,
  • Fouzi Harrou,
  • Ying Sun

摘要

Accurate defect detection in steel plates ensures safety, prevents damage, and reduces repair costs. This study explores ensemble machine learning techniques for defect detection, addressing the challenges of multivariate, imbalanced data with relevant and irrelevant features. Six ensemble models are evaluated on full and reduced feature sets (using Principal Component Analysis (PCA)). To mitigate data imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is applied. A publicly available dataset with 1941 instances, 34 columns, and 27 independent variables is used, comprising seven fault types. The data was split into 80% for training and 20% for testing. Results show that SMOTE significantly enhances classification accuracy, with Stacked Ensemble and XGBoost achieving near-perfect accuracies of 0.9916 and 0.9915, respectively. Additionally, SHapley Additive exPlanations (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) are used to interpret model predictions. These findings underscore the effectiveness of combining SMOTE with ensemble models in improving automated defect detection systems for steel plates.