The Research And Design Of Telecom Customer Credit Model Based On Stacking Model Fusion

In recent years,with the popularization of mobile devices and the continuous development of the communications industry,the needs of telecommunication customers have become increasingly sophisticated and complex.At present,there are a large number of user information records in the database of telecom operators.However,these data information are huge in quantity and complex in type.The method of manual analysis can't process and use the valid data well,and does not exert the data value that should be available to the effective data in the database.Through the work of this paper,telecom operators can initialize the user's credit rating,establish a credit rating system,reduce the cost of user credit research work,and enhance the telecom operator's brand image and profits.The work results and research results of this paper are as follows:1.Deal with large amounts of data in the telecommunications industry databaseThe number of database data tables provided by the telecommunications industry is very large,about a thousand database tables.First,the valid tables are selected and merged by the user ID.Then,the missing rows of data are supplemented or deleted by default,while natural language features are converted to numeric labels that can be used for classification.In addition,the data is trained and the feature values used by the model are filtered according to the contribution of the feature values.Finally,training sample selection is based on the Hard Mining.2.Build a telecom user credit model based on the integration of Stacking modelRandom Forests,GBDT and DAG SVMs were used as secondary classifiers to classify samples.Then,through the Stacking method,the prediction results of the three sub-classifiers are converted into meta-feature vectors for constructing a meta-classifier.Finally,the fusion classification algorithm obtained by the above three algorithms for model fusion constitutes a complete telecommunications user credit degree model,which is used to evaluate the user's initial creditworthiness.3.Evaluate model effectsAfter the model is constructed,it is compared with other integrative learning methods.In addition,the model will be compared with the model fusion method such as voting method and weight method.