| The estimation of production time is an essential part of the manufacturing domain, allowing companies to optimize their production plan and meet the dates required by the customers. In the last years, there have been several approaches that use Machine Learning (ML) to predict the time needed to finish production orders. In this paper, we use the CRISP-DM methodology and Automated Machine Learning (AutoML) to address production time prediction for a Portuguese contract manufacturing company that produces metal containers. We performed three CRISP-DM iterations using real data provided by the company related to production orders and production operations. We compared four open-source modern AutoML technologies to predict production time across the three iterations: AutoGluon, H2O AutoML, rminer, and TPOT. Overall, the best results were achieved in the third CRISP-DM iteration by the H2O AutoML tool, which obtained an average error of 3.03 days. The obtained results suggest that the inclusion of data about individual manufacturing operations is useful for improving production time for the entire production order. |
*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.