近几年，随着移动互联网和货运物流行业的深度融合，网络货运平台应运而生，一定程度上解决了过去车货信息匹配难的问题。国内外目前关于网络货运平台的研究主要围绕在比如网络货运的服务模式、路径规划、平台的系统价格研究、定价策略，关于网络货运成交的影响却相对较少。探究对数字化货运平台成交的影响因素，对完善我国数字化货运平台、促进全社会物流资源更高效率运转具有非常重要的意义。本文的目的是通过构建货物成交概率的模型，分析出货物成交有影响的相关因素，并依据相关因素对货物是否成交进行合理预测。本文基于某网络货运平台，收集了900条货物信息数据，包括所需车长、货物重量、运输里程、首选车型、运输价格等，利用Logistic回归、分类决策树、随机森林方法建立货物成交模型，并通过逐步回归以及树模型变量重要性得分优化Logistic的设计向量来改进模型以提高模型的预测准确率。实证研究表明，货物在是否有一口价的两种情况下分别受到不同的因素影响，有一口价的货物受所需车长、是否当日可装、运输里程、运输价格等因素影响较大，没有一口价的货物主要受到所需车长、货物重量、货主好评率和可选车型数等因素影响较大。通过比较模型的AIC值以及反复实验下模型的平均准确率和准确率的方差，结果表明使用优化设计向量后Logistic回归模型预测是否成交效果更好。这也从侧面印证了利用决策树模型改进逻辑回归模型的设计向量这一思路是有效的，也是本文的一大创新点。

目前，关于网络货运平台成交概率的实证研究相对较少，一定意义上讲本文的研究是具有探索性的，由于缺乏公开的网络货运平台成交数据，本文只能设计实验收集数据，样本的数据质量和实验设计的严谨性在未来都有待完善，所以研究结论的普遍性和准确性还有待考证。未来的研究可以从增加更多的特征变量、统计模型的改进、实验本身响应变量的改善比如将具体成交时间分箱，进行多方面的实证研究。

In recent years, with the deep integration of mobile Internet and freight logistics industry, NTOCC platform came into being, which solved the problem of difficult matching of vehicle and cargo information in the past to a certain extent. At present, the research on NTOCC platform at home and abroad mainly focuses on the service mode, path planning, systematic price research and pricing strategy of online freight, but the impact on NTOCC transaction is relatively small. Exploring the factors affecting the transaction of digital freight platform is of great significance to improve China's digital freight platform and promote the more efficient operation of logistics resources in the whole society. The purpose of this paper is to analyze the relevant factors affecting the transaction of goods by constructing the model of transaction probability of goods, and reasonably predict whether the transaction of goods is completed according to the relevant factors. Based on a NTOCC platform, this paper collects 900 pieces of cargo information data, including the required vehicle length, cargo weight, transportation mileage, preferred vehicle model, transportation price, etc. the cargo transaction model is established by using logistic regression, classification decision tree and random forest method, and the logistic design vector is optimized by step-by-step regression and tree model variable importance score to improve the model, so as to improve the prediction accuracy of the model. Empirical research shows that the two cases of whether goods have a buy it now price are affected by different factors. The goods with a buy it now price are greatly affected by the required vehicle length, whether they can be loaded on the same day, transportation mileage, transportation price and other factors. The goods without a buy it now price are mainly affected by the required vehicle length, cargo weight, praise rate of the owner and the number of optional models. By comparing the AIC value of the model and the average accuracy and accuracy variance of the model under repeated experiments, the results show that the logistic regression model is better to predict whether the transaction is better after using the optimized design vector.This also proves that the idea of using decision tree model to improve the design vector of logistic regression model is effective, and it is also a major innovation of this paper.

At present, there are relatively few empirical studies on the transaction probability of NTOCC platform. In a sense, the research of this paper is exploratory. Due to the lack of public transaction data of NTOCC platform, this paper can only design experiments to collect data. The data quality of samples and the preciseness of experimental design need to be improved in the future, so the universality and accuracy of the research conclusions need to be verified. Future research can be carried out by adding more characteristic variables, improving the statistical model and improving the response variables of the experiment itself, such as dividing the specific transaction time into boxes.