Prediction And Analysis Of Domestic Used Car Prices Based On Ensemble Learning

Nowadays,as the used car market involves various industries,predicting the price of used car in the market has become a hot research topic,as it requires a lot of effort in analysis,also the knowledge of domain experts,and collecting a lot of different car attribute to make reliable and accurate predictions.Therefore,this dissertation regards actual domestic used car data as the object of analysis,and studies how to apply ensemble learning to used car data to establish a used car price evaluation model that is in line with actual applications.This paper first draws out the advantages and disadvantages of the development of the domestic market from the development history of the foreign second-hand car market,and determines the main research ideas of the thesis.After referring to the current main valuation methods,the ensemble learning method is finally used to construct a used car valuation model.The input characteristics of the model are related attributes of used cars: used time,model,brand,body type,fuel type,gearbox,engine power,mileage,whether it is damaged,etc.The relevant variables have been desensitized,and Information from more than one hundred thousand vehicles has been used for modeling.All models have been modeled and evaluated in Python software.In the research process,the following ensemble learning models were considered: RandomForest,XGBoost,Light GBM.These models were merged using Stacking,Blending and linear weighting.Then,this paper introduces the best choice for each model in MAE.The modeling results show that the Light GBM model has the fastest training speed among the three models and the highest prediction accuracy.The MAE of the model is 592.43,which is 3.5% and 4.7%smaller than the errors of the other two models.On this basis,through model fusion,it is found that the prediction results after using Stacking and linear weighting are better,and the fusion MAE are 579.82 and 579.28,which is 2.2% smaller than the Light GBM model error.The effect of model fusion is significantly higher than the effect of a single model.Base on the model established in this paper,it can speed up the establishment of used car price evaluation systems for used car merchants and financial and insurance industries,promote the optimization of a fair and just used car price evaluation system,and promote the improvement of a reasonable and standardized second-hand car trading market system.