Mobile Network Traffic Prediction Using MLP, MLPWD, and SVM

Abstract

Mobile networks are critical for today's social mobility and the Internet. More and more people are subscribing to mobile networks, which has led to substantial demands. The network operators need to find ways of meeting the huge demands. Since mobile network resources, such as spectrum, are expensive, there is a need for efficient management of network resources as well as finding a way to predict future use for network management and planning. Network planning is crucial for network operators to provide services that are both cost effective and have high degree of quality of service (QoS). The aim of this research is to apply data analysis techniques to support network operators to maximize the resource usage for network operators, that is, to prevent both under-provisioning and over-provisioning. Therefore, this paper investigates the prediction accuracy of machine learning techniques -- Multi-Layer Perceptron (MLP), Multi-Layer Perceptron with Weight Decay (MLPWD), and Support Vector Machines (SVM) -- using a dataset from a commercial trial mobile network. The experimental results show that SVM outperforms MLP and MLPWD in predicting the multidimensionality of the real-life network traffic data, while MLPWD has better accuracy in predicting the unidimensional data. Our experimental results can help network operators predict future demands and facilitate provisioning and placement of mobile network resources for effective resource management.