PREDICTING AND ANALYZING SHIPPING TIME USING ENSEMBLE TREE MODELS WITH SHAPLEY ADDITIVE EXPLANATIONS

Panupong Wanjantuk; Ruth Banomyong

Proceedings of ICLT 2024

PREDICTING AND ANALYZING SHIPPING TIME USING ENSEMBLE TREE MODELS WITH SHAPLEY ADDITIVE EXPLANATIONS

Panupong Wanjantuk; Ruth Banomyong

Department of Computer Engineering, Khon Kaen University, Thailand; Center of Excellence in Connectivity, Thammasat Business School, Thammasat University, Thailand

International Conference on Logistics & Transport 2024, Seoul, South Korea, pp. 120-128

Download PDF | View interactive page

Abstract

Purpose: This study aims to develop a predictive model for shipping times using advanced machine learning techniques, specifically ensemble learning models, to enhance accuracy and reliability in supply chain management. Additionally, the study seeks to interpret the model's predictions to understand the influence of various factors on shipping durations, providing actionable insights for optimizing logistics operations. Design/methodology/approach: The research utilizes a dataset comprising 5,114 rows of historical shipment records. Data preprocessing included one-hot encoding for categorical variables, scaling numerical features, and addressing imbalanced distributions. Seven ensemble tree models were employed. The models were evaluated using cross-validation and various evaluation metrics. SHAP (SHapley Additive exPlanations) was used to interpret the best-performing model, providing insights into feature importance and interactions. Findings: The CatBoost model demonstrated the highest accuracy in predicting shipping times, followed by Random Forest and LightGBM. SHAP analysis revealed that normalized shipment charges, processing days, and turnaround time thresholds were the most significant features influencing shipping times. Interaction plots highlighted the complex dependencies between features. Research Limitation: The study's limitations include the quality and completeness of the dataset, which can affect model performance. Additionally, the models' predictive power may diminish when applied to significantly different future conditions or new shipping routes not represented in the training data. Practical Implication: The predictive models developed in this study can be integrated into logistics management systems to provide real-time shipping time estimates and insights, enhancing inventory management, reducing costs, and improving customer satisfaction. The interpretability provided by SHAP helps logistics managers optimize processes, resource allocatio

Keywords

Shipping Time Prediction; Ensemble Learning Models; SHAP; Supply Chain Management; Predictive Analytics

Citation

Panupong Wanjantuk; Ruth Banomyong (2024). PREDICTING AND ANALYZING SHIPPING TIME USING ENSEMBLE TREE MODELS WITH SHAPLEY ADDITIVE EXPLANATIONS. Proceedings of the International Conference on Logistics & Transport (ICLT 2024), Seoul, South Korea, pp. 120-128.