Study On Influencing Factors And Forecasting Models Of Household Used Car Value Preservation

Today,the used car market in China is developing rapidly,and the transaction volume of used cars continues to grow.In 2022,a total of 16 million cars were traded nationwide.Among them,affected by the insufficient supply of new cars,the used car quasi-new car market attention increased significantly.The emergence of the Internet,e-commerce and social networks has promoted the second-hand trading market to transform into an online trading platform more quickly.The economic form extended by the Internet+has gradually become familiar to people.However,there is a lack of unified evaluation system and scientific evaluation standards in the second-hand car market.Currently,the evaluation methods of second-hand cars mainly rely on the current market price method and the traditional evaluation methods such as referring to the replacement cost method,which are very dependent on the subjective experience of evaluators and lack of objectivity and scientific nature.Therefore,to explore a more objective and accurate rapid evaluation method for our country’s second-hand car market transparent development has important practical significance.In terms of the arrangement of the content of this thesis,first of all,refer to the relevant literature to understand the traditional asset evaluation methods and mathematical model methods;Then,sample data were obtained by crawler combined with vehicle base,and the random forest model,LightGBM model and decision tree model were established respectively for feature correlation analysis of sample data.Moreover,the constructed models were compared and analyzed from multiple dimensions such as goodness of fit,prediction accuracy and stability.Finally,the integrated learning combination model with the best forecasting effect is obtained based on the Stacking thought.The experimental results show that the comparison model generalization performance:the Stacking model is the strongest,the LightGBM model is the second,the random forest model is the third,and the decision tree model is the weakest.The following conclusions are drawn:(1)Compared with the single learner model,ensemble learning is easier to process category variables,less susceptible to multicollinearity,and has the advantages of less noise data impact and accurate prediction.(2)By comparing the algorithms represented by the two ideas of ensemble learning,LightGBM is superior to random forest in model prediction ability and construction performance.(3)Compared with the other two ideas of integrated learning,the Stacking thought can combine the advantages of a single model and achieve the optimal effect of stacking by constantly reducing deviations.Analyze the influencing factors of the second-hand car preservation rate through the model prediction results:the most important factor is the vehicle use dimension,which is respectively the age of the vehicle and the number of kilometers.The next most important factor is the basic attributes of the vehicle,which are brand,vehicle series,model,model and vehicle emission standards.The general important factors are the car body structure and power factors,respectively for the body structure,whether the engine is turbocharged,transmission and fuel type.The results of this study can be used for reference by all interested parties,including consumers,dealers,automobile manufacturers and relevant government departments.