Leveraging Contact Pattern to Predict Future Contact Pattern in Mobile Networks

With advances in the Internet and mobile technology, and decreasing cost of mobile devices, large scale pervasive networks are now ubiquitous in solving many earlier service limitations. Here, the challenge lies in its underlying temporal graph. It introduces technical limitations in efficient routing, maximal coverage with minimal latency, data offloading, to effective dissemination over mobile networks or mobility induced dynamic networks. Efficient solution to these interrelated problems lies in the novel prediction strategies for most accurate future contacts (links or interactions), their future contact time etc. In contrast to the existing strategies that consider either network structure or regular pattern and periodic nature of contacts, we propose a novel stochastic Poisson process model (variants of cascaded non-homogeneous Poisson process) that employ multi-recurrent, dependent contact pattern as its basis. We predict number of contacts relative to a node and over all nodes in any future interval, future contact time over a user and a pair of users. Finally, we validate our model with a widely used empirical data set from mobile network, and compare our model with doubly recurrent and homogeneous Poisson process model to conclude the superiority of our prediction model.