MDI and PI XGBoost regression-based methods: regional best pricing prediction for logistics services

Abstract

The logistics industry in Indonesia, with PT Pos Indonesia as the dominant player, is confronted with intense price competition. The challenge lies in establishing the most favorable price for regional logistics services in every region, with the aim of gaining a competitive edge and augmenting revenue. This intricate task encompasses local market conditions, competition, customer preferences, operational costs, and economic factors. To address this complexity, this study proposes the utilization of machine learning for price prediction. The price prediction model devised incorporates the extreme gradient boosting regression (XGBR), support vector machine (SVM), random forest, and logistics regression algorithms. This research contributes to the field by employing mean decrease in impurity (MDI) and permutation importance (PI) to elucidate how machine learning models facilitate optimal price predictions. The findings of this study can assist company management in enhancing their comprehension of how to make informed pricing decisions. The test results demonstrate values of 0.001, 0.005, 0.458, 0.009, and 0.9998. By employing machine learning techniques and explanatory models, PT Pos Indonesia can more accurately determine optimal prices in each region, bolster profits, and effectively compete in the expanding regional market.