A data-driven approach to predict the value and key features of collectible cars

• Autor(a) principal: Pires, Pedro Miguel Geraldes
• Data de Publicação: 2020
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10071/21876
• topic: Data mining; Collectible cars market; Alternative investments; Emotional assets; Coronavirus; Mineração de dados; Mercado de carros colecionáveis; Investimentos alternativos; Ativos emocionais

In our study, a Data Mining approach is proposed to predict the selling price of collectible cars at auction and determine which vehicle's characteristics influence this value. RM Sotheby's, a prestigious auction house, allowed us over 30,000 vehicles in order to build our data set. The use of different models allowed us to analyze a large data set with 18 features. In order to determine which model is most suitable for our study, 11 Data Mining models were compared using 4 metrics (MAE, NMAE, RAE and RMSPE), the 11 models were tested on our data set using a rolling window scheme. "Xgboost", a decision tree model, presented the best results (RMSPE = 12.69%). In addition, a method of extracting knowledge from sensitivity analysis was applied, which allowed us to determine the key features in the sales price (Brand.continent, Car.age, km.categorized and Model.identifier). In addition, a comparison with traditional investments was made and it was used to analyze the impact of coronavirus. Our results demonstrated a superior performance when compared to iron, FTSE MIB and FTSE 100. In the case of coronavirus, there was a significant drop in volume and sales since the beginning of coronavirus, although the average price per car has increased, which does not allow us to guarantee that there was an impact in this market. A more detailed analysis to 25 different models was conducted. From these, 9 saw an appreciation in price while the remaining devalued. In total, those 25 models had an overall devaluation of 5%.