Data Mining Methods, Models and Solutions for Big Data Cases in Telecommunication Industry

Abstract

AbstractThis paper is concentrated on the applications of main data mining tools, methods, models and technologies for solving the basic tasks of data processing for telecommunication industry. The main forecasting task in this industry is to predict the class and volume of services needed for the subscribers as well as for predicting the capacity of all needed engineering equipment. It is proposed to develop regression and forecasting models based upon the Facebook Prophet module to take into account seasonal effects in data and to compare the results with the ones received by using the LSTM network. The classification task was related to the problem of churn prediction. The authors identified such promising methods as decision trees, random forest, logistic regression, neural networks, support vector machines and gradient boosting to solve problems of subscribers classification by their certain preferences and services, as well as the tendency to outflow. The dynamic approach based on dynamic models of survival theory for churn time prediction is proposed. Next the task of forecasting the volume and class of services which subscribers are going to use in roaming is solved. The results of using regression models and data mining methods were also shown. All the methods proposed were compared and evaluated by necessary statistical characteristics, and interpretation of the model application results for practical solutions were proposed. Finally in the paper the client-service information technology with all needed functionality for solving all these tasks is proposed.KeywordsBig DataData miningChurn predictionRegression modelsFacebook ProphetGradient boostingLSTM networkDynamic risk assessmentMobile InternetTelecommunication company