Applying machine learning approaches to price prediction

Machine learning based solutions are widely applied to price prediction problems. Although linear estimators such as generalized linear models or linear regressors are generally preferred to estimate the price of a product in the market using its properties, nonlinear models that are more flexible are not often used. In this study, a model for price estimation developed by using the past and current sales data of approximately 14 million cars, considering the technical specifications and financial indicators at the date of sale, and the performance level compared with the common algorithms. In the light of this analysis, the performance of the model that predicts the price of a car at a certain date examined. In determining the value of the cars, besides its technical characteristics, financial indicators ignored in previous studies have been taken into account. Learning curves, overfitting or underfitting situations and the error values also examined by testing the model on subsets of the data set with different characteristics.