Ensemble Method of Multiple Decision Trees with Crisp and Fuzzy Discretization for Axial Surface Roughness Prediction

Abstract

AISI 1045 is a steel used to make engine components for motor vehicles and aircraft. The quality of AISI 1045 is influenced by its surface roughness, both axial and tangential. The prediction of AISI 1045 surface roughness aims to produce quality products in a shorter time and at a lower cost. The axial surface roughness is in the form of certain grades according to ISO. This method is a prediction technique that combines several single prediction models to obtain better prediction results. In this work, the ensemble method is built using a voting system from an odd number of single prediction models of the decision trees. The proposed Single Model consists of one decision tree model with crisp discretization (DT1) and three models with fuzzy discretization (DT2, DT3, and DT4). The research data were obtained through experiments measuring the axial surface roughness of AISI 1045 steel using a wet machining system by considering cutting speed, feed motion, axial depth of cut, and tangential surface roughness. The study’s results indicate that not all proposed ensemble models are built to have better performance than single prediction models. Of the four proposed single prediction models, only one model has an accuracy above 80%, namely the decision tree model with fuzzy discretization using a combination of linear-trapezoidal fuzzy membership functions (DT4 model). The model performance based on accuracy, recall, precision, F1-score, and AUC is 80.73%, 48.53%, 73.47%, 58.44%, and 67.72%, respectively. For the four ensemble models formed from the combination of three Single decision tree models, only the combination of DT1, DT2, and DT3 does not perform better than the Single model. The other three ensemble methods have better accuracy, recall, and AUC than the performance of all proposed Single models with values of 81.33 - 82.67%, 51.62 - 55.19%, and 69.04 - 71.02%, respectively.