Análise preditiva de preços de automóveis usados
Date
2024
Journal Title
Journal ISSN
Volume Title
Publisher
IPCB. EST
Abstract
Este estudo investiga a análise preditiva de preços de automóveis usados, destacando a importância de prever o valor de mercado dos veículos com base em características como marca, quilometragem, ano de fabrico e tipo de combustível. A pesquisa explora diversas técnicas de machine learning para gerar previsões mais precisas, incluindo regressão linear, florestas aleatórias (random forest) e redes neuronais artificiais.
A metodologia adotada baseia-se na análise de datasets do mercado automóvel do Reino Unido, examinando padrões de desvalorização e identificando os principais fatores que influenciam a variação dos preços ao longo do tempo. O estudo evidencia que a abordagem tradicional de regressão linear apresenta limitações na modelagem da depreciação dos veículos, sendo a curva em S um modelo mais adequado para representar esse comportamento de forma realista.
Os resultados obtidos indicam que modelos baseados em random forest oferecem maior precisão preditiva em comparação com métodos estatísticos convencionais. A análise gráfica, incluindo heatmaps e regressões, reforça a importância da quilometragem, marca e ano de fabrico como variáveis determinantes na precificação de veículos usados.
Por fim, o estudo propõe o desenvolvimento de uma aplicação funcional baseada nos modelos explorados, permitindo prever preços futuros de automóveis usados. Essa ferramenta visa auxiliar consumidores, concessionárias e seguradoras a tomarem decisões mais informadas no mercado de segunda mão, reduzindo incertezas e aumentando a eficiência das transações comerciais.
Abstract : This study investigates the predictive analysis of used car prices, emphasizing the importance of estimating vehicle market value based on factors such as brand, mileage, year of manufacture, and fuel type. The research explores various machine learning techniques to generate more accurate predictions, including linear regression, random forests, and artificial neural networks. The methodology is based on the analysis of datasets from the UK used car market, examining depreciation patterns and identifying the key factors influencing price variations over time. The study highlights that the traditional linear regression approach has limitations in modeling vehicle depreciation, while the S-curve proves to be a more suitable model for realistically representing this behavior. The results indicate that random forest-based models provide higher predictive accuracy compared to conventional statistical methods. Graphical analysis, including heatmaps and regression models, reinforces the significance of mileage, brand, and year of manufacture as key variables in used vehicle pricing. Finally, the study proposes the development of a functional application based on the explored models, enabling the prediction of future used car prices. This tool aims to assist consumers, dealerships, and insurance companies in making more informed decisions in the second-hand market, reducing uncertainties, and increasing transaction efficiency.
Abstract : This study investigates the predictive analysis of used car prices, emphasizing the importance of estimating vehicle market value based on factors such as brand, mileage, year of manufacture, and fuel type. The research explores various machine learning techniques to generate more accurate predictions, including linear regression, random forests, and artificial neural networks. The methodology is based on the analysis of datasets from the UK used car market, examining depreciation patterns and identifying the key factors influencing price variations over time. The study highlights that the traditional linear regression approach has limitations in modeling vehicle depreciation, while the S-curve proves to be a more suitable model for realistically representing this behavior. The results indicate that random forest-based models provide higher predictive accuracy compared to conventional statistical methods. Graphical analysis, including heatmaps and regression models, reinforces the significance of mileage, brand, and year of manufacture as key variables in used vehicle pricing. Finally, the study proposes the development of a functional application based on the explored models, enabling the prediction of future used car prices. This tool aims to assist consumers, dealerships, and insurance companies in making more informed decisions in the second-hand market, reducing uncertainties, and increasing transaction efficiency.
Description
Keywords
Análise preditiva, Machine learning, Preços de automóveis usados, Regressão, Florestas aleatórias, Curva em S, Predictive analysis, Used car prices, Regression, Random forests, S-curve
Citation
MENDES, João Luís Lebre Ramos Antunes ; SILVA, Pedro Miguel da Fonseca e (2024) - Análise preditiva de preços de automóveis usados. Castelo Branco : IPCB. EST. Relatório do Trabalho de Fim de Curso de Engenharia Informática.