Prediction On Truck-to-cargo Matching Of Freight Platform Based On Stacking Model

The traditional truck-to-cargo matching market,with opaque prices,lack of integrity,and asymmetry in information,makes the freight market middlemen occupy the right to control prices.With the changes in the industry situation,the Didi Taxi model has made the car-cargo matching platform realize that car owners and cargo owners can be quickly matched through the Internet,thereby solving problems such as information asymmetry.Therefore,the use of effective methods can increase the success rate of vehicle-to-goods matching,help companies improve push efficiency,rationally allocate resources,and improve the imbalance between supply and demand.Firstly,The paper analyzes and summarizes the current research status of domestic and foreign road freight matching.Then,compared with other integrated learning algorithms,Stacking integrates models that have good performance on certain specific problems,and cross-validation methods can make The final generated model has higher prediction accuracy,and it is determined that the prediction of vehicle and cargo matching based on the Stacking fusion model method is selected.According to the experimental background of this paper,the problem of binary classification of unbalanced data,support vector machines,random forests,GBDT and XGBoost are selected as the primary learners of the Stacking model,and the technical principles of these models are introduced.In the context of truck-to-cargo matching,through qualitative analysis and comprehensive consideration of all aspects,the factors affecting freight matching used in predictive modeling are determined.And through the data provided by the P platform,based on the actual business scenario,the relevant features were analyzed and selected,the optimal feature subset was screened out,and some highly relevant features were constructed,such as path similarity.Based on the above five models,this paper uses the vehicle-cargo matching of the P platform as an example to carry out modeling and prediction.Through comparison,it is found that the Stacking fusion model performs well compared to a single model,and Stacking performs better in the face of unbalanced samples.excellent.This paper proposes that the integrated learning method can provide reference and reference value for other online freight platform's vehicle-to-cargo matching prediction research.Finally,taking the P platform as an example,based on the user characteristics of data analysis,suggestions are made for the development of the network freight platform.