Analysis Of Second＿hand Car Users’ Transaction Intention Model Based On Machine Learning Algorithm

Based on the digital needs of China’s second-hand car industry,the development of second-hand car e-commerce platform is increasingly mature.The trading patterns of second-hand automobile e-commerce platforms are gradually becoming more and more colorful,among which C2 B pattern,as an important trading pattern,plays an important role in the development of second-hand automobile market.Based on the C2 B trading model of the used car e-commerce platform,it is of great significance to study the user behavior from the user’s perspective,grasp the user behavior intention,and promote the transformation of the online trading of the used car,for the development of the C2 B trading model and even the automobile market.This thesis makes use of the user behavior data in the C2 B mode of a used car e-commerce platform to explore the influence of user behavior on users’ trading intention.Firstly,the messy actual engineering data are cleaned and transformed,including the imputation of missing values,the categorization of numerical variables and the dummy coding of categorical variables.Secondly,for the unbalanced data of the categories,resampling processing is conducted from the data level to compare the prediction effect of the original data with the random undersampling,random oversampling and the original data processed by the sampling algorithm on different models.The recall rate and AUC value are taken as the evaluation criterion of the models,and the unbalanced data are finally processed by random undersampling.Finally,Logistic regression,random forest and XGBoost algorithms are used to construct the transaction intention model.In Logistic regression,L1 regularization is added to prevent overfitting merger for variable selection.The random forest and XGBoost algorithms are used with the variable selection method based on the feature importance of under-sampling to select some variables and fit the model.By evaluating the recall rate and AUC value of each model,it is concluded that the prediction effect of random forest is the best.From the perspective of the influence of variables on user intention,variables such as first-level source,variable information filling,user identity information,car age and mileage have an impact on user transaction intention.