Forecasting China Future MNP by Deep Learning

Abstract

The objective of this study is to find a most accurate way to forecast the future of the Mobile Number Portability (MNP) users in Mainland China. We propose a simplified MNP AD System that suits Chinese situation. MNP is an optional value-added-service through which customers can retain their assigned mobile telephone numbers but change their subscriptions from one mobile network operator to another. The service has been on the move for more than 19 years over 80 countries in the world except Mainland China, even with Macau and Hong Kong. Consequently, relatively few data from China are available, and the insufficiency of training data poses a forecasting challenge. Sixteen machine learning methods including contemporary deep learning algorithms are used in an attempt of forecasting the future MNP of China; however, the prediction accuracy is acceptable only for large dataset. When the dataset is small, univariable time series forecasting fail to predict with reliability. By introducing more factors that are related to the forecasting objective to the dataset (turning it multi-variable), the accuracy improves with error rate drops. The accuracy is found to further rise after removing some irrelevant factors. Finally propose some recommendations, simplified process, and a centralized MNP AD system with less human work that is applicable to Mainland China are proposed. The system is easy for government to control the porting and better forecast as business intelligence use.