Thesis Defence: A Data-Driven Supplier Selection and Order Allocation Optimization Framework Under Tariff and Demand Uncertainty

Marzieh Eskandari Khanghahi, supervised by Dr. Babak Tosarkani & Dr. Abbas Milani, will defend their thesis titled “A Data-Driven Supplier Selection and Order Allocation Optimization Framework Under Tariff and Demand Uncertainty” in partial fulfillment of the requirements for the degree of Master of Applied Science in Mechanical Engineering.

An abstract for Marzieh Eskandari Khanghahi’s thesis is included below.

Defences are open to all members of the campus community as well as the general public. This defence will be offered in hybrid format.  Registration is not required to attend in person; however, please email babak.tosarkani@ubc.ca to receive the Zoom link for this defence.

Abstract

Global supply chains are increasingly exposed to demand uncertainty, tariff fluctuations, geopolitical disruptions, and supplier capacity limitations. These challenges make Supplier Selection and Order Allocation (SSOA) decisions more complex, especially in multi-product and multi-period settings where procurement, production, transportation, inventory, and shortage decisions are closely interdependent. Traditional cost-minimization approaches may become unreliable when demand patterns shift, or tariff policies change the relative attractiveness of suppliers across regions.

This thesis develops a tariff-aware supplier selection and order allocation framework under demand uncertainty using a Machine Learning (ML)–robust optimization approach. The proposed framework integrates demand forecasting, residual-based data-driven uncertainty modelling, and mixed-integer linear programming (MILP) to support procurement, production planning, and inventory management decisions. Statistical, conventional ML, and deep learning (DL) forecasting models, including Seasonal Autoregressive Integrated Moving Average with Exogenous Regressors (SARIMAX), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN)-based models, are evaluated to estimate future product demand. Forecast errors are then used to construct data-driven uncertainty sets based on Principal Component Analysis (PCA) and Kernel Density Estimation (KDE), allowing the model to capture empirical correlations, asymmetry, and non-Gaussian behaviour in demand deviations. Finally, these uncertainty sets are incorporated into a robust optimization model that accounts for tariff-adjusted procurement costs, supplier capacities, quantity discounts, transportation decisions, warehouse limits, overtime, and shortage penalties.

The novelty of this thesis lies in integrating tariff-aware global sourcing, ML-based demand forecasting, residual-based PCA-KDE uncertainty construction, and robust supplier selection and order allocation within a unified decision-support framework. Numerical experiments based on a representative global electronics supply chain show that tariff changes significantly affect sourcing allocations, total network cost, transportation choices, and shortage levels. The results indicated that forecast-based planning can reduce nominal planning costs, while robust solutions provide stronger protection against demand underestimation and shortage risk.