| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 5.05 MB | Adobe PDF |
Advisor(s)
Abstract(s)
Atualmente, a previsão de valores de indicadores de gestão, tais como vendas, procura,
rentabilidade, etc., desempenha um papel importantíssimo no mundo empresarial, contribuindo,
de forma impactante, nas decisões estratégicas das organizações. As previsões tradicionais,
baseadas em fórmulas simples ou na opinião de peritos, apenas, estão a ficar obsoletas e a ser
substituídas por modelos de previsão automatizados e mais modernos. Isto porque, muitas vezes,
existe complexidade nas tendências dos dados disponíveis que não são corretamente detetados
pelos métodos mais simples.
O presente trabalho teve como objetivo principal a aplicação de algoritmos de previsão baseados
em modelos estatísticos e em métodos de aprendizagem automática para melhorar a precisão das
previsões de matrículas futuras da Toyota Caetano Portugal. Os trabalhos iniciaram com uma
revisão sobre métodos de previsão, seguindo-se uma análise exploratória exaustiva dos dados
disponíveis e a modelação via modelos estatísticos e modelos de aprendizagem automática da
procura e das vendas de automóveis da Toyota Caetano Portugal. A seleção do melhor método foi
efetuada com base em métricas adequadas.
Através da análise dos resultados, observou-se que o modelo baseado em redes neuronais foi o
que produziu erros mais baixos. Além disso, as previsões obtidas com esse modelo apresentaram
valores próximos dos valores reais e são mais precisas do que aquelas anteriormente realizadas
pela Toyota Caetano Portugal.
Como contribuição, este trabalho demonstra a viabilidade da aplicação de algoritmos de previsão
no setor automóvel de forma a melhorar a eficácia das projeções de matrículas. Estes resultados
têm uma importância prática tremenda, pois ajuda a empresa a tomar decisões mais corretas, e a
realizar o plano de matrículas e os orçamentos com maior confiança.
Currently, the effectiveness of forecasts plays an important role in the business world, making a significant contribution to strategic decisions. For this reason, traditional forecasts, based on simple formulas, are becoming obsolete and are being replaced by more modern and automated forecasting models. This is because complexities in data trends are often not accurately detected. The purpose of this work is to apply forecasting algorithms based on machine learning to enhance the accuracy of Toyota Caetano Portugal's future registrations. In this academic work, in addition to adopting a quantitative approach through descriptive data analysis, a standard process model (CRISP-DM) with an adapted structure for the case study was applied, and various forecasting algorithms were explored. Through results analysis, we observed that the neural network model achieved remarkably low errors, standing out as the one that best learned the complex patterns and seasonality from the data. Additionally, it generated close predictions to the actual values and more accurate than those previously made by Toyota Caetano Portugal's. These results highlight the importance of considering seasonality when dealing with temporal data. As a contribution, this work demonstrates the feasibility of applying forecasting algorithms in the automotive sector to improve the registrations projections effectiveness. These acquired results hold tremendous practical importance, as they help the company to make more correct decisions, and to carry out the registration plan and budgets with greater confidence.
Currently, the effectiveness of forecasts plays an important role in the business world, making a significant contribution to strategic decisions. For this reason, traditional forecasts, based on simple formulas, are becoming obsolete and are being replaced by more modern and automated forecasting models. This is because complexities in data trends are often not accurately detected. The purpose of this work is to apply forecasting algorithms based on machine learning to enhance the accuracy of Toyota Caetano Portugal's future registrations. In this academic work, in addition to adopting a quantitative approach through descriptive data analysis, a standard process model (CRISP-DM) with an adapted structure for the case study was applied, and various forecasting algorithms were explored. Through results analysis, we observed that the neural network model achieved remarkably low errors, standing out as the one that best learned the complex patterns and seasonality from the data. Additionally, it generated close predictions to the actual values and more accurate than those previously made by Toyota Caetano Portugal's. These results highlight the importance of considering seasonality when dealing with temporal data. As a contribution, this work demonstrates the feasibility of applying forecasting algorithms in the automotive sector to improve the registrations projections effectiveness. These acquired results hold tremendous practical importance, as they help the company to make more correct decisions, and to carry out the registration plan and budgets with greater confidence.
Description
Keywords
Car Market Registrations Forecast Data Mining Machine Learning Forecasting Algorithms