Optimizing Random Forest Models for Early Detection of Defects in Steel

Abstract

In the manufacturing sector, steel plate defects are a severe issue that may result in significant losses for a company's finances and image. The purpose of this study is to evaluate how well three machine learning algorithms detect steel plate flaws. The accuracy, area under the ROC curve (ROC-AUC), and Log-Loss of the method were used to assess its performance using a dataset that was downloaded from www.kaggle.com. Based on the findings, the Random-Forest algorithm performed best overall, having the lowest Log-Loss of 0.9327, an accuracy of 0.6722, and an AUC value of 0.9222. Research using other algorithms is still very open to be carried out to get better results. Research utilizing other algorithms is still very much open to be conducted in order to get better outcomes.