COMPARISON OF PREDICTIVE MODELS FOR AN APPLIED DISTRIBUTOR PALLET LOADING PROBLEM

This paper is motivated by the operations of an Italian ceramic production company that deals daily with the distributor’s pallet loading problem. The industry is interested in a fast and accurate approach to this problem. We propose a hybrid approach, combining constructive heuristics with different machine learning models: XGBoost, LightGBM, and Random Forest. Using data from the Italian industry, the models are trained, validated, and compared with the software PackVol. The proposed models showed superior accuracy and computing time performance, indicating that the hybrid approach is effective even with different machine learning models.