Prediction And Analysis Of Used Car Transaction Prices

Although the used car trading market has developed rapidly and gradually prospered in recent years,there are still many problems.Especially for the evaluation of used car transaction prices,there are differences in the evaluation methods of transaction prices among platforms and individuals,resulting in many difficulties in the transaction process.Therefore,in order to reduce the random pricing behavior of various platforms during transactions,it is particularly necessary to make an accurate estimate of the transaction price of used cars,so as to make the used car transaction market more standardized.This paper firstly analyzes the current situation of the used car trading market at home and abroad,and then determines the research methods and ideas.There are external factors and internal factors that affect price fluctuations.External factors are macro factors,such as policy factors,economic factors,etc.,while internal factors are micro factors,mainly related to the nature of the vehicle itself,such as brand,mileage,and usage time.Since external factors are uncontrollable and cannot be intervened by people,internal factors directly determine the transaction price of used cars under the condition that external factors remain unchanged.Therefore,this paper starts with the internal factors that affect used cars,and analyzes and processes the relevant variables before modeling.In view of some current mainstream machine learning methods,this paper first selects three single models,Light GBM,random forest,and extreme random forest to predict and analyze the transaction price of used cars.The data selected is the price data of used cars on the Tianchi Big Data Competition website,the variables are model,brand,kilometers traveled,etc.The three single models were modeled and evaluated by using Python scripts,and the MAEs obtained were 699.75,622.66,and 573.69,respectively.Then,the Stacking fusion model method is used,and Light GBM,random forest,and extreme random forest are used as the base model.The second layer uses logistic regression,and the MAE after fusion is 588.62.By comparison,it is found that the Stacking fusion model method has a good prediction effect.Therefore,the fusion model proposed in this paper has certain practical application value.