【作者】 刘颖；

【导师】 谢田法；

【作者基本信息】 北京工业大学 ， 应用统计（专业学位）， 2019， 硕士

【摘要】 近年来,随着我国经济持续快速增长,国内汽车保有量也在稳步增长。由于消费观念的转变,越来越多的人选择购买二手车作为出行工具,二手车市场需求不断增加,规模也在不断扩大。由于“互联网+”战略的快速实施,我国二手车交易市场进入新的发展阶段,各项数据均表明我国二手车市场发展势头良好,拥有非常广阔的发展空间。然而,与国外成熟的二手车市场相比,我国二手车市场起步较晚,仍然存在一些问题。在二手车交易中,无论对二手车购买用户还是经销商来说,对二手车的保值率进行合理的估值都是至关重要的。因此,快速并且准确地评估二手车的保值率,不仅能加速二手车市场流通的速度,还能保证买卖双方利益达到最大化。目前二手车的估价还主要是由评估师根据自身的经验进行,评估的随意性较大。利用数据挖掘或者其他方法来建立二手车相关评估模型的研究还处在初步探索阶段,至今还没能找到一种操作方便,准确率高的二手车评估模型。本文将借鉴国外车辆价格评估的经验,采用数据挖掘方法预测二手车保值率,建立合理的二手车保值率预测模型。本文基于网络爬虫方法从汽车之家网站收集了2万余条二手车源信息,选取包括品牌、上牌时间、表显里程等可能影响二手车价格的25个因素。首先用传统特征选择、Boruta算法和Lasso回归三种方法对变量进行初步筛选,总共选取了15个变量用于后续的建模。然后,主要选取机器学习方法中的随机森林回归算法和GBDT算法来建立二手车的保值率预测模型。最后运用十折交叉验证法计算两种模型的预测误差,对比模型的误差发现,随机森林的预测误差较小,预测精度较高,预测效果较好。结果表明运用随机森林回归方法建立二手车保值率的预测模型是值得推广的。从重要性排序结果来看,上牌时间、品牌、表显里程以及车系等是影响二手车保值率的主要因素。更多还原

【Abstract】 In recent years,China’s economy has grown rapidly,and domestic car ownership is also growing steadily.Due to the change of consumption concept,more and more people choose to buy second-hand cars as travel tools.The demand for second-hand car market is increasing and its market scale is also expanding.Due to the rapid implementation of the "Internet +" strategy,the second-hand car trading market in China has entered a new stage of development.All the data indicate that the development of second-hand car market in China is good and has a very broad space for development.However,compared with the second-hand car market in foreign countries,the second-hand car market in China started late,and there are still some problems.In the second-hand car transaction,it is crucial to make a reasonable valuation of the second-hand car’s hedging rate for both used car buyers and dealers.Therefore,quickly and accurately assessing the value-added rate of used cars not only accelerates the speed of the used car market,but also maximizes the interests of buyers and sellers.At present,the evaluation of used cars is mainly carried out by the evaluators based on their own experience,and the evaluation is somewhat random.The research of data mining or other methods to establish used car evaluation models are still in the preliminary stage of exploration,so far,a second-hand vehicle evaluation model with convenient operation and high accuracy has not been found.This thesis will learn from the experience of foreign vehicle price assessment,use data mining method to predict the second-hand car hedging rate,and establish a reasonable prediction model of the second-hand car hedging rate.Based on the method of web crawler,more than 20,000 second-hand vehicle source information was collected from the automobile home website,and 25 key factors which may affect the price of second-hand cars were selected,including brand name,time of registration,and mileage.Firstly,the variables were initially screened by traditional feature selection,Boruta algorithm and Lasso regression.A total of 15 variables are selected for further analysis.Then,the random forest regression algorithm and GBDT algorithm in machine learning method are mainly selected to establish the prediction model of the preservation rate of used cars.Finally,the ten-fold cross-validation method is used to calculate the prediction results of the two models.By comparing the errors of the two models,it is found that the prediction errors of random forests are smaller,the prediction accuracy is higher and the prediction effect is better.The results indicate that it is worth popularizing to use the random forest regression method to establish the prediction model of the preservation rate of used cars.Judging from the importance variables,the time of branding,brand,mileage and car system are the main factors affecting the value of used car.更多还原