Explainable artificial intelligence-enabled intertwined logistics analytics with application in poultry vaccine distribution in the Northwest of Iran.

Intertwining supply chains integrates the corresponding networks across several intersection points, such as suppliers, manufacturers, and transporters, resulting in higher viability and efficiency. Nonetheless, no formulation has been proposed for intertwined logistics planning. In addition, interpretable machine learning models, such as multiple linear regression models, may not be an accurate metamodel of vehicle assignment-routing models. This paper contributes to intertwined logistics planning by proposing a mixed-integer linear mathematical model for vehicle assignment and routing in the multi-logistic, multi-depot, multi-vehicle distribution network planning problem. In addition, a framework for global sensitivity analysis of mathematical models based on eXplainable Artificial Intelligence (XAI) is proposed. The proposed mathematical model is solved in a case study of poultry vaccine distribution involving two logistics, four depots, 10 customers, and eight vehicles. Moreover, the proposed framework is applied by training several AI algorithms, such as linear regression, random forest, XGBoost, CatBoost, LightGBM, ANN, RNN, 1D-CNN, and LSTM networks, using 80% and 20% of a clean dataset of size 222 for training and test, respectively. Results showed that the designed intertwined distribution network is efficient because the shipping, travelling, and opportunity costs are 80%, 16%, and 4%, respectively, due to six within-depot and two layoff vehicle assignment. Furthermore, fleet stress tests showed that the designed intertwined distribution networks is resilient under fleet stress test, such as two vehicles lost, using between logistics vehicle sharing strategy. In addition, the results of applying XAI-based global sensitivity analysis revealed that shipping cost has the most significant impact on total costs variability, followed by demand.