A Research Of Customer Churn Prediction Within Mobile Internet Background

In recent years,with the rapid development of mobile Internet,more and more industries begin to apply mobile Internet technology to provide diversified mobile services.And further they expand scenarios of customers' activities.With the help of mobile Internet,e-commerce,accommodation,catering,tourism services etc.get a development.More and more industries are focusing on mobile service.While these industries now face many serious problems,including fierce competition, homogenous products or services and uneven quality of service etc.These problems directly lead to heavy loss of customers among the industries within mobile Internet background.In view of this phenomenon,this study preprocesses the data from user behavior,excavates the factors that reflect user preferences and will of chum in daily behaviors,achieves effective and accurate customer churn prediction and helps industries achieve better and better development through correlative suggestions and measures.The problem of customer churn prediction comes down to be a classification problem.At present,there are many classification algorithms commonly used in practice,but they can't be of high accuracy,high efficiency and strong interpreting ability simultaneously.So this paper selects a more suitable method—the Xgboost classification algorithm.But there would be some problems to use the Xgboost directly in this paper:setting a sparse matrix without filling the missing value will influence the data processing thereafter;adjusting the built-in parameters of Xgboost to handle imbalanced data will cause the prediction probability of no significance;too many variables will affect the classification efficiency etc.That makes data preprocessing a key prerequisite.In order to make the algorithm more feasible and effective in this paper,here tries to use the corresponding preprocessing methods to solve the problems above:Using MICE for the missing value imputation.The modified Borderline-SMOTE method is proposed to solve the imbalanced problem.Different feature selection algorithms are applied to reduce the features and obtain the core variables.The conclusions of this study are:preprocessing of the original data before building the classification model has great impacts on the model results for the data samples are very large.Fill orders or not,customer activity&customer traffic and customer conversion rate are important attributes that influence customer churn in the mobile hotel reservation market.The conclusions and suggestions will help improve the customer relationship management in related markets through user behavior information.The data preprocessing and model ideas in this paper can be the reference for customer churn analysis among the related industries,which has meaning both in theory and realistic.