Research On Second-hand Car Price Forecast Based On Feature Optimization Combination SVM

At present,China's second-hand car market is not perfect,there are still many problems.Especially for the evaluation of second-hand car value,institutions and individuals have different evaluation prices of second-hand cars,and their differences cause great difficulties for the transaction of second-hand cars.With the development of the era,second-hand car trading scale is increasing,in view of the second-hand car trading market price too high or too low,and in the process of mediation platform is wanton for bad behavior,need from the data index of second-hand car itself,to accurately estimate of used cars,to protect the legitimate rights and interests of consumers,reduce the price of consumer losses,promote the smooth development of the used-car market.The evaluation and prediction of the second-hand car price is helpful to standardize the industry standards of the second-hand car sales platform,solve the rationality of the second-hand car price evaluation,make the seller sell the satisfactory price and the buyer buy the valuable goods.In this paper,the second-hand website everyone used car price assessment,using the web crawler technology to collect data,then the data preprocessing,data conversion,fill the missing value,an analysis of the correlation data index,the index characteristics and descriptive analysis used car price data,it is concluded that brands parameters of vehicles,analysis used car price distribution.Then all the raw data using support vector machines SVM to forecast the used-car prices,on the basis of the model tuning effect optimize the index data,using the characteristics of the engineering,respectively using PCA principal component analysis,random forests and gradient feature selection ascension tree GBDT,more will choose the information expression,importance of larger characteristic index and the SVM model to predict,so as to optimize model prediction effect,make them more accurate evaluation of second-hand car prices.The results show that there is a relationship between the data indicators and the second-hand car prices,and there are differences between the prices and configurations of second-hand cars of different brands.For the three prediction models with feature optimization,they are all better than the traditional support vector machine.The SVM model combined with GBDT feature optimization has the best prediction effect on second-hand car price,with the minimum error between the predicted value and the real value,and the prediction result of the SVM model withrandom forest feature optimization is the most stable.For midrange car price projections,GBDT characteristics optimization in combination with the SVM model to predict the best effect,difference in 0-3 or so,but for high-end cars,GBDT feature optimization combining the SVM prediction model and the real value has certain deviation,so suggest the consumers at the high end need to look at other configuration information when buying a used car,to make more rational judgment.Finally,the paper puts forward Suggestions from three perspectives: national policy,second-hand car trading platform and consumers.