Wavelet transform processing for cellular traffic prediction in machine learning networks

Abstract

The ability for cellular operators to closely predict the network traffic volume at various locations can be very important for their resource management and dynamic network control including offloading. This work investigate the analysis of the spatial-temporal information of cellular traffic flow and the prediction of cell-station traffic volumes. Based on the integration of K-means clustering, Elman Neural Network (Elman-NN), and wavelet decomposition methods, we characterize the performance comparison of traffic volume prediction. We tested our wavelet decomposition based machine learning approach using the real traffic data recorded at a district in a big city and demonstrated the gain over traditional approaches.