Jiahui Chen Article first published online: 2018 Abstract
This research aims to apply a series of classical machine learning algorithms based on decision trees (Decision Tree, Adaboosting, Bagging, Random Forest) to verify the ten-fold cross-validation of the steel plate fault data. The source of the data set was the Research Center of Sciences of Communication in Italy and has been used two times by M Buscema when it is provided [15, 16]. The data set includes 7 different types of steel plate faults: Pastry, Z_Scratch, K_Scatch, Stains, Dirtiness, Bumps, and Other Faults. It is found that the Bagging algorithm outperforms the other methods and achieves 96.30% and 90% accuracy on the training and testing set, respectively. This will allow us to find abnormalities on the surface of the steel plate timely and reduce losses. Based on these algorithms, we can cooperate with iron and steel practitioners to design more appropriate algorithms to achieve higher recognition accuracy in the future. |