Sales forecast using meta-learning model as complementary to lean 4.0

• Ano de defesa: 2024
• Autor(a) principal: Leo, Cicero Gomes
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Palavras-chave em Português: FFORMA; Lean 4.0; Meta-learning; Forecast; Industria 4.0; Meta- aprendizagem; Método de previsão; Algoritmo; Produção intelectual; Meta-aprendizado; Previsão
• Link de acesso: http://app.uff.br/riuff/handle/1/32594

The discoveries about Lean 4.0 are far from exhausted, but only some studies still empirically expose the application of meta-learning models in this context. Therefore, this study aims to recognize the complementarity between prediction models based on meta-learning algorithms and Lean 4.0 and empirically evaluate their performance. For this purpose, a systematic literature review was conducted using the Systematic Search Flow (SSF) method, recognizing the strengths and weaknesses of meta-learning models in relation to complementarity with Lean 4.0. It was possible to identify that companies more committed to Lean are more likely to implement Industry 4.0 technologies, just as greater engagement from senior leadership increases the chance of success in implementing such technologies. The FFORMA model (Feature-based FORecast Model Averaging) was selected based on previously established criteria and applied to data from a company in the metallurgical segment in Southeast Brazil. Time series augmentation using autoregressive Gaussian mixture (MAR) models were used to train the meta-learning model, aiming to improve the model learning experience and, therefore, improve forecasting performance. The results related to the model's prediction error performance showed that the combined prediction was inferior compared to the individual prediction methods used in the combination. Out of the 19-time series, the combined model performed well in only two-time series. However, applying the meta-learning model proved interesting when a financial analysis of the impact on the inventory was conducted. Significant opportunities for reducing excesses by up to 94% of the inventory were observed, in addition to the relationship with the strengths and weaknesses in the literature review. This research aims to provide a solution for organizations with problems associated with sales forecasting, pricing strategy, storage space allocation, inventory management, distribution, and replacement plans in the metallurgical sector. They need sales forecasting as an integral part of the decisionmaking process and contribute as a theoretical reference related to Lean 4.0 and meta-learning models.