Movement Prediction in Wireless Networks Using Mobility Traces

Wireless user-mobility prediction has been investigated from various angles to improve network performance. Student populations in campuses, pedestrian and vehicular movement in urban areas, etc have been studied by cell phone and mobility management researchers to address issues in Quality of Service (QoS), seamless session handoffs, etc. Access to information such as user movement times, direction, speed, etc provides an opportunity for networks to efficiently manage resources to satisfy user needs. Towards this goal, we propose a generic framework to approach the problem of mobility prediction using Hidden Markov Models (HMM). This method can be used to modd hidden parameters in the models. We propose a way to extract user movement information from a real dataset, train a HMM using this data and make predictions using the HMM. This model can successfully predict long sequences of a mobile user's path from observed sequences and also uses successive sequences of observed data to train its learning parameters to enhance prediction accuracy. Furthermore, we show that this model is very generic and can be suited to make predictions using the same information from the perspective of the access point or the mobile node.