Subscriber Location Prediction: A Neural Network Approach

Abstract

Location forecasting of subscriber is critical in cellular network. Since many issues such as handoff, blocking probability, user prediction, etc., are greatly influenced by this location movement of mobile user. In this work, a comprehensive study of different artificial neural techniques and its wireless application, i.e., user forecasting, is presented. Three types of typical neural networks, namely back propagation (BP), Legendre (LN) and radial basis function (RBF), are extensively studied, investigated and analysed in the paper. The location user data used are the hourly mean mobile user data collected at sites in Chandrasekharpur, Bhubaneswar area. The performance is evaluated based on three metrics, namely training accurate, testing accuracy and processing time. The random user where the movement is frequent then conventional algorithms like multi-layer perceptron (MLP), radial basis function (RBF) do not outperform the Legendre neural network (LNN). For the best performance, the nonlinear neural network selected also depends on the type of collected mobile user data. The performance matrices such as processing time, training accuracy and testing time obtained from the simulation results that outperform, i.e., 0.006239 s, 84 and 81% than conventional neural algorithms. This indicates the implemented algorithm is a single robust and reliable that forecasts the location of a roaming user in a wireless network.