Stochastic optimization and machine learning applied in the demand forecast, allocation and distribution of vaccines between Brazilian states.

• Autor(a) principal: Lopes, Juliano Marçal
• Data de Publicação: 2022
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da USP
• Texto Completo: https://www.teses.usp.br/teses/disponiveis/3/3143/tde-05092022-095859/
• topic: Aprendizado computacional; Demand forecast; Demanda (Previsão); Immunobiological demand; Otimização estocástica; Stochastic optimization; Vaccines; Vacinas (Distribuição)

Advances in research and development have resulted in the emergence of many new vaccines in recent decades. However, the distribution of vaccines and the fight against vaccine-preventable diseases is still a challenge for chain managers. The vaccine supply chain typically has limited budgets, difficulty controlling product temperatures, poor inventory management, and lack of protocol for high demand and uncertain situations. Mismanagement of the vaccine supply chain can lead to a disease outbreak or, at worst, a pandemic. Fortunately, a large number of vaccine supply chain challenges such as optimal dose allocation, improving vaccination strategy and inventory management, among others, can be improved through optimization approaches. Given this scenario, the objective of this work is to propose methods to reduce costs in the chain. This was done through the creation of a machine learning model to forecast demand and a stochastic optimization model to improve the distribution of immunobiologicals among Brazilian states. The models presented here, despite considering the Brazilian scenario, have the potential to have their applications extended to the vaccine supply chain in other countries. To carry out this work, first visits were carried out in five Brazilian states to understand and map the processes of the vaccine distribution chain of the Ministry of Health. This mapping allowed the solutions proposed here to be elaborated taking into account the current scenario of the chain. The developed machine learning model encompasses the use of Gradient Boosting and Random Forest Regressor techniques, and its results are used as input data for the proposed optimization model. The stochastic optimization model considers the uncertain demand of three scenarios. The results of the study show that the machine learning model presents a demand forecast with errors significantly lower than those that the chain currently presents. Furthermore, the results of the optimization model help decision makers with a suggestion of the number of doses that should be sent to each state in each of the months of the considered period, thus reducing the chance of vaccine shortages.