Portfolio Forecasting Research On Target Positioning Of Wealth Management Products

The research and evaluation of the classification algorithm has been a hot research topic.In the previous studies,we found that most of the content is to study different algorithms for the same classification method,very few of the content is to study data mining classification method and discrete choice model of the traditional combination used in the same issue.Therefore,in this paper,we not only choose the data mining algorithms,also chose discrete choice models to solve the problem that for customers to choose financial products.What is more,in the use of data mining algorithms,we choose different data mining classification algorithms to do a comparison.In the process of concrete implementation,the problems in the choice of financial products for customers,we use the existing research results,and consider the maturity of the theory of the classification method.In the data mining part,we choose the decision tree classification method and Bias classification method.In the decision tree classification method,the decision tree C5.0 algorithm is selected.In the Bias classification method,the naive Bias classification algorithm is chosen.In the discrete choice model,considering the fact that our problem is a two part classification problem,we choose the traditional standard logit model to solve it.We use C5.0 decision tree classification algorithm model,Naive Bayesian classification model and the standard logit model based on the empirical analysis.The empirical results based on three methods,using the combination forecast theory,combination of local optimal prediction models,in comparison of the prediction accuracy,our mainly index is the prediction accuracy of the model.In the empirical of the single model,we use the Bank data in the UCI database set,random data sets will be in accordance with the ratio of 7:3 is divided into training set and test set,in order to eliminate the influence of random data distribution,we use the same training data to set up model,use the same test data to test the results of the model.And we find that the accuracy of C5.0 decision tree classification model,Naive Bayesian model and standard logit model in identifying the customers of financial products are the 88.43%,89.48% and 87.51%.From the results of the classification of the model,we can see that the accuracy of the three models is more than 80%,so we can say the precision of the models is high.Among them,the results of the naive Bayes model is the best,the accuracy is as high as 89.48%.Then,after we introduce the weighted least square,weighted arithmetic average,weighted inverse variance method,weighted mean square inverse and the weighted simple method,the five commonly used method for calculating weight model combination.With the results of C5.0 decision tree model,Naive Bayesian model and standard logit model,and based on the results of the calculation of weighted least squares method,the weighted arithmetic average method,weighted inverse variance method,weighted mean square inverse and the weighted simple method,five new combination models are obtained.Finally,after we get the five models,we found that the one which use the weighted least squares method,the weight of standard logit model is negative,not in line with the actual situation,therefore,we think the least squares weighted combination model should be adopted.The accuracy of the model weighed simple method less than single model,so we also adopt the result.In comparison with the accuracy of the weighted arithmetic average model,we found that the error probability of weighted arithmetic average is 10.09%,less than the 10.48% error probability of weighted inverse variance model.The weighted arithmetic average weighted model has a higher degree of accuracy.Therefore,we choose the arithmetic average weighted model as our optimal model with an accuracy of 89.91%,and the accuracy of the combined model is improved by a factor of 0.43% compared with a single model.