Predictive modelling for shipment pricing in supply chain management

Abstract

The current research paper develops a predictive model of freight cost behaviour in global supply-chain events, with the use of machine-learning model, using a sample of 10,324 transaction data of the SCMS Delivery History sample. The model examines how the physical characteristics of shipment, financial values, the terms of the contract and the logistics environment combine to affect freight spending. The comprehensive analytical pipeline was used, which includes data cleaning, feature engineering, categorical consolidation, dimensionality reduction, and supervised learning using the Linear Regression, Random Forest, and XGBoost models. A set of more than 13,000 one-hot encode predictors was then reduced by low-variance filtering after preprocessing; the data was then divided into 75 percent training data and 25 percent test data. The findings suggest that XGBoost outperforms all control models and reaches an RMSE of USD6826, MAE of USD3344 and R 2 of 0.79 on the test set despite the presence of extremely skewed cost data.