MSFS: Multiple Spatio-Temporal Scales Traffic Forecasting in Mobile Cellular Network

With peta-bytes of data that are continuously collected about various aspects of the mobile networks, one of the main challenges when dealing with such data is performing accurate predictions in order to address a broad class of application problems, ranging from mobile network optimization to preventive maintenance. To this end, time series prediction has been widely addressed by statistics community. However, the performance of time series prediction is seriously affected by spatio-temporal context. In this paper, with spatio-temporal hierarchical constructions, we aggregate time series and propose a Multi-Scales Forecasting System (MSFS) to predict time series of different granularities. From the aspect of statistics, three traditional methods, Seasonal ARIMA (SARIMA), Hidden Markov Model (HMM) and Wavelet Neural Network (WNN), are applied to investigate accuracy and complexity of different spatio-temporal time series prediction. MSFS is evaluated with a real-word 3G dataset, and the experimental results show that time series in different spatio-temporal context have disparate characters. Through comparing the statistical results of each spatio-temporal scale, we analyze which granularity can be predicted and give a recommend proposal of forecasting method. Although this scheme is based on statistical results, it also opens possibilities for the development of more efficient traffic engineering and anomaly detection tools, which will result in financial gains from better network resource management.