Transfer Learning for Multi Step Resource Utilization Prediction

Long term accurate and efficient resource utilization predictions are of vital importance for the future generation of mobile wireless networks. By anticipating network resources demand, the operator can get a clearer view of how service quality will impact the user experience and trigger the appropriate optimization function. In this paper we evaluate the use of deep and transfer learning algorithms for multi step resource utilization prediction in one or multiple time intervals in the future by using minimal information from a given cell. We propose the use of LSTMs with transfer learning to address the existing challenges in terms of scalability and data storage of current implementations. We carry out extensive experiments on a dataset collected from a live customer LTE network. Our approach is proven to achieve state of the art results with RMSE below 12 for a 4 hours forecast. We then analyze the complexity overhead and factors that led to the achieved performance.