This paper conducts model training and testing for estimating the price of used cars. In the second-hand car market, the residual value is one of the most important reasons for the second-hand car circulation. Because much of the way used cars are traded is traditional used car appraisers rely on their experience with used cars to determine the value of the car. At present, there is no comprehensive evaluation model based on objective factors of second-hand car characteristics. This study is based on the establishment of an objective evaluation of the effectiveness of the price prediction model and the objective factors affecting the second-hand car. In this research, the author wants to replace traditional evaluation methods with deep learning methods. The code written in this study includes the training of the data preprocessing model, flexibly uses NumPy to perform precision calculations, and uses Sklearn to delete highly relevant features for feature screening. The non-linear isomap algorithm is used for dimensionality reduction to keep the data at the same latitude, and finally the SVM algorithm is used for model training. Very unsatisfactory results have been obtained by training the model. The train accuracy and test accuracy were 22.5% and 14.3%, respectively. Through the analysis of the model, it may be that the dimensionality reduction of the isomap algorithm deleted the required influencing factors, resulting in the low training accuracy of the entire model.