Comparison of Defective Casting Product Classification Results Using the K-Nearest Neighbors Algorithm

Abstract

This study compares the accuracy of K-Nearest Neighbors (KNN) and Naive Bayes algorithms in detecting defects in impeller products. Using a dataset of impeller images, we applied preprocessing, feature extraction, and selection techniques. The models were assessed using metrics such as precision, accuracy, F1-score, recall. and with KNN achieving 98.11% accuracy and Naive Bayes 85.38%. The t-SNE visualization confirmed distinct clustering of defective and non-defective products. Our findings suggest that KNN is more reliable for defect detection in industrial applications. These results provide valuable insights for implementing effective machine learning models in manufacturing quality control.