Designing a resilient humanitarian supply chain by considering viability under uncertainty: A machine learning embedded approach

Humanitarian supply chains (HSCs) play a crucial role in mitigating the impacts of natural disasters and preventing humanitarian crises. Designing resilient HSCs is critically important to ensure effective recovery and long-term sustainability during and after such events. This study addresses the design of resilient HSCs with viability consideration under known-unknown demand and capacity uncertainties by formulating a two-stage stochastic programming model. To solve this problem and achieve high-quality solutions, three solution approaches are developed and compared. The first approach introduces risk aversion into a genetic algorithm (GA) through chance constraints, termed GA with chance constraints (GAC). The other two approaches integrate the Random Forest (RF) algorithm with GAC, employing incremental learning (GACRFI) and non-incremental learning (GACRFNI). To evaluate the performance of these algorithms and provide insights into designing a resilient HSC, a full factorial design of experiments (DoE) is established using controllable factors. Problems are generated for three cases, each of which corresponds to a distinct disruption and ripple effect severity degree. Computational analysis shows that integrating the machine learning algorithm into the GA yields superior results across all risk level settings, leading to a win–win situation for all stakeholders in HSCs. This study provides valuable insights for designing resilient HSCs that ensure both short-term recovery and long-term sustainability by considering viability under varying risk levels and severity degrees.