Research On Mining Potential Users Of ETC Based On User Profile

With the rapid development of Internet technology and the advent of the era of big data,enterprises benefit from the development of network and technology,collect a large number of user data and information.However,in the face of massive user data,enterprises are unable to effectively screen out valuable user information,presenting the situation of low data utilization rate and low data asset conversion rate.Since the government work report in 2019 proposed to strengthen the promotion of ETC,major banks and emerging Internet finance companies have joined the competition to seize the users of ETC.Due to the homogeneity of ETC products,we need to combine marketing strategy and product service combination to form differentiated competitiveness.In addition,there is a lack of recognition and experience accumulation of ETC in the enterprise,and the response to the market is slow.Enterprises urgently need to achieve the goal of user data mining,accurate promotion and improving the efficiency of user management.Based on the above problems,this paper makes an in-depth discussion and Research on the mining of potential users of etc,mainly including the following four aspects:This paper selects the data set of ETC potential users mining in a domestic artificial intelligence modeling contest in 2019 as the research object.Firstly,the data set is cleaned and coded,and then the processing method of unbalanced data is determined through comparative experiments,so that the proportion of positive and negative samples in the data set tends to be balanced,and the processed training data set is generated.In addition,the ID feature processing method is proposed and verified.Experiments show that the method can remove the sample duplication and retain the number of samples,which improves the efficiency and accuracy of the model.Subsequently,AUC was the first mock exam basis,supplemented by F1,Precision and Recall indicators to evaluate the effects of 6 single models(LOGISTIC regression,Random Forest,XGBoost,Light GBM,Deep Neural Network,CNN).Finally,according to single models,we propose to integrate the single model with Stacking method and form a ETC user portrait combination model based on multi model ensemble learning.Compared withsingle models and other multi model ensemble methods,the ETC user portrait combination model based on multi model ensemble learning has higher AUC value than other models used in this paper,which verifies the validity and performance of the model.