Used Car Price Prediction Using CatBoost Regression Model

In today's auto industry, determining resale value for a car after buying is essential. A vehicle’s value is influenced by many factors including age; amount driven; fuel used; type of transmission; ownership history; and manufacturer, but are most significantly influenced by dealer assessments based on both personal experience and emissions testing conducted at the time of sale and used as a basis for valuations. When using the methods that assess a car's value to assist in valuation by a dealership when determining retail value for a vehicle, there may be significant differences in the quality of those assessments and therefore based on dealer appraisal and their experience, can result in different price assessment results. This paper presents a machine learning-based model of predicting the value of previously owned vehicles using CatBoost Regression, whereby historical automobile data was collected from publicly available data sources. Before applying the model, preprocessing techniques were employed for cleaning and improving overall quality of the data; as well as removing duplicate records and/or records with missing values, as well as dealing with outliers. The CatBoost model has a superior ability to predict fair values of market price by considering relevant attributes for a given automobile with higher accuracy than what can be achieved using traditional automobile dealer pricing methods by way of model comparison against performance metrics defined by each relative metric of performance which are: MAE, RMSE, and R2 value based; the CatBoost algorithm proved capable of successfully and accurately predicting values for the category based variable type. The resulting model from this research serves as an objective and clear way for potential buyers/sellers of a vehicle to estimate prices for vehicles being bought/sold via online market.