كنفرانس بين المللی مهندسی صنايع و سيستم ها

صفحه اصلی / دوازدهمین كنفرانس بين المللی مهندسی صنايع و سيستم ها

An Explainable Tuned Machine Learning Framework for Supplier Lead Time Prediction and Decision Support

نویسندگان :

Mansoureh Naderipour¹ Mohammad Hossein Zangiabadi²

1- شهید باهنر کرمان 2- شهید باهنر کرمان

کلمات کلیدی :

lead time prediction،machine learning،XGBoost،particle swarm optimization،explainable AI،SHAP،supply chain decision support

چکیده :

Accurate prediction of supplier delivery lead time is essential for inventory planning, yet most existing studies pursue prediction accuracy alone and leave the drivers of lead time unexplained. This paper proposes a two-part framework for supplier lead-time prediction under a currency-allocation-controlled import regime. First, four baseline regressors (Linear Regression, Decision Tree, Random Forest, and XGBoost) are trained on 72,186 historical import orders and compared under identical settings; the best-performing model is then refined with Particle Swarm Optimization (PSO). Second, SHapley Additive exPlanations (SHAP) are embedded as a structural stage of the framework, rather than a post-hoc analysis, to attribute each prediction to its underlying business factors at both the portfolio and individual-order level. On the held-out test set, the PSO-tuned XGBoost model achieves a mean absolute error of 6.80 days and R² = 0.962, improving on all baselines. The explainability analysis identifies the time required for foreign-currency allocation, a financial and regulatory step that precedes supplier payment, as the dominant driver of lead time, exceeding the contribution of conventional logistics factors such as distance and shipment weight. A scenario analysis translates this finding into concrete managerial numbers, showing that reducing the currency-allocation delay from 40 to 5 days shortens predicted lead time by over 100 days. The resulting Optimized Model → Explainable Optimized Model → Decision Support pipeline supports differentiated safety-stock policy, supplier evaluation, and process-improvement decisions.