Impacto do uso de técnicas de aprendizado de máquina, big data e variáveis causais na gestão de demanda
Carregando...
Arquivos
Citações na Scopus
Tipo de produção
Dissertação
Data
2022
Autores
Silva, Janderson Lopes
Orientador
Sampaio, Mauro
Periódico
Título da Revista
ISSN da Revista
Título de Volume
Citação
SILVA, Janderson Lopes. Impacto do uso de técnicas de aprendizado de máquina, big data e variáveis causais na gestão de demanda. 2022. 72 p. Dissertação (Mestrado em Engenharia Mecânica) - Centro Universitário FEI, São Bernardo do Campo, 2022 Disponível em: https://doi.org/10.31414/EM.2022.D.131509.
Texto completo (DOI)
Palavras-chave
aprendizado de máquina,previsão de demanda,cadeia de suprimentos (gerenciamento)
Resumo
Este trabalho de dissertação tem como principal objetivo avaliar o efeito da utilização de algoritmos de aprendizado de máquina e Big Data na acuracidade de previsão de demanda, além de analisar a influência das variáveis causais internas e externas. Para alcançar esse objetivo, realizou-se uma revisão sistemática da literatura com os principais conceitos de gestão de demanda para avaliar os principais métodos, abordagens e fatores relevantes para construção do modelo de planejamento de demanda. Além disso, estudou-se também os principais algoritmos de aprendizado de máquina e Big Data, que pudessem melhorar no desempenho do modelo. Após, elaborou-se um procedimento metodológico de estudo de caso em uma empresa do segmento de agronegócio, com coleta e análise de dados da demanda planejada e realizada, dos últimos cinco anos e, na sequência, realizou-se análises quantitativas para validar as hipóteses e análises qualitativas através de entrevistas com os especialistas da área. Ambas análises foram importantes para elaboração final do modelo. Como resultado, obteve-se que a utilização das técnicas de aprendizado de máquina e Big Data, melhorou em 37% a acuracidade de previsão de demanda em relação ao método qualitativo de ajuste de julgamento utilizado pela empresa. Além disso, o modelo proposto em 100% dos casos, utilizou de métodos mais robustos e sofisticados de aprendizado de máquina, do que utilizar métodos tradicionais de séries temporais simples. Outro fator importante é que as variáveis causais internas contribuíram em 100% dos casos da família de fungicidas para melhoria do planejamento de demanda e, em contrapartida, as variáveis causas externas apoiou em 50% dos casos, para melhoria desse modelo. Conclui-se, então, que as principais contribuições deste trabalho de pesquisa foram: avaliar o desempenho da acuracidade de previsão de demanda comparando métodos clássicos de séries temporais, com métodos de aprendizado de máquina; verificar as influências das variáveis causais internas e externas nos modelos de previsão de demanda; utilizar software de planejamento de demanda com modelos mais robustos, de fácil utilização e com um rápido tempo de processamento; e por fim, estruturar uma revisão sistemática da literatura, com as pesquisas mais recentes dos estudiosos da área de gestão de demanda e Supply Chain 4.0, com intuito de utilizar as melhores pesquisas praticadas no mercado
The main objective of this dissertation is to evaluate the effect of using machine learning algorithms and Big Data on the accuracy of demand forecasting, in addition to analyzing the influence of internal and external causal variables. To achieve this objective, a systematic review of the literature was carried out with the main concepts of demand chain management to evaluate the main methods, approaches and relevant factors for the construction of the demand planning model. In addition, the main machine learning algorithms and Big Data were also studied, which could improve the performance of the model. Afterwards, a methodological procedure of a case study was elaborated in a company of the agribusiness segment, with collection and analysis of data of the planned and carried out demand, of the last five years and, subsequently, quantitative analyzes were carried out to validate the hypotheses and qualitative analysis through interviews with experts in the field. Both analyzes were important for the final elaboration of the model. As a result, it was found that the use of machine learning and Big Data techniques improved the accuracy of demand forecasting by 37% in relation to the qualitative method of judgment adjustment used by the company. In addition, the proposed model in 100% of cases used more robust and sophisticated machine learning methods than using traditional simple time series methods. Another important factor is that the internal causal variables contributed in 100% of the cases of the fungicide family to improve demand planning and, on the other hand, the external causes variables supported in 50% of the cases to improve this model. It is concluded, then, that the main contributions of this research work were: to evaluate the performance of demand forecast accuracy comparing classical time series methods with machine learning methods; verify the influences of internal and external causal variables on demand forecasting models; use demand planning software with more robust models, easy to use and with a fast processing time; and finally, to structure a systematic review of the literature, with the most recent research from scholars in the area of demand management and Supply Chain 4.0, in order to use the best research practiced in the market
The main objective of this dissertation is to evaluate the effect of using machine learning algorithms and Big Data on the accuracy of demand forecasting, in addition to analyzing the influence of internal and external causal variables. To achieve this objective, a systematic review of the literature was carried out with the main concepts of demand chain management to evaluate the main methods, approaches and relevant factors for the construction of the demand planning model. In addition, the main machine learning algorithms and Big Data were also studied, which could improve the performance of the model. Afterwards, a methodological procedure of a case study was elaborated in a company of the agribusiness segment, with collection and analysis of data of the planned and carried out demand, of the last five years and, subsequently, quantitative analyzes were carried out to validate the hypotheses and qualitative analysis through interviews with experts in the field. Both analyzes were important for the final elaboration of the model. As a result, it was found that the use of machine learning and Big Data techniques improved the accuracy of demand forecasting by 37% in relation to the qualitative method of judgment adjustment used by the company. In addition, the proposed model in 100% of cases used more robust and sophisticated machine learning methods than using traditional simple time series methods. Another important factor is that the internal causal variables contributed in 100% of the cases of the fungicide family to improve demand planning and, on the other hand, the external causes variables supported in 50% of the cases to improve this model. It is concluded, then, that the main contributions of this research work were: to evaluate the performance of demand forecast accuracy comparing classical time series methods with machine learning methods; verify the influences of internal and external causal variables on demand forecasting models; use demand planning software with more robust models, easy to use and with a fast processing time; and finally, to structure a systematic review of the literature, with the most recent research from scholars in the area of demand management and Supply Chain 4.0, in order to use the best research practiced in the market