Abstract:

Purpose: The main purpose of the study was determined as modeling, interpreting and comparing the model performances of the variables with machine learning algorithms. Based on the original data in 2018-2019 from a company operating in the manufacturing sector, variables that are effective on the Overall Equipment Effectiveness (OEE) score in the ordinal scale obtained by the simulation study are used. Methodology: Decision trees (CART, CHAID), logistic regression (LogR) and artificial neural networks (ANN) were used in modeling the OEE score. The performance values of the established models were calculated according to the criteria of “sensitivity”, “specificity”, “precision” and “accuracy”. While interpreting the models, percentages were used for decision trees and ANN results, and odds ratio was used for LogR. Findings: In the models, "hour", "production", "experience" and "lost meter" variables were examined on OEE score. By comparing the performance criteria, it was determined that the algorithm that gave the best results was ordinal LogR. It has been determined that those with low production and less experience have a “lower" OEE score, and those with less lost meters have a higher chance of getting a “higher" OEE score. Originality: OEE, which is modeled as a continuous in the literature, was made categorical and the companies were able to determine their own positions by determining the difference between the classes. Thus, companies will be able to move up to the next category faster by determining their category and changing the variables in order of importance in the selected model. The lack of models in the literature that analyze categorical OEE with machine learning has been determined as the originality of this study.

Keywords: