A novel network traffic prediction method based on a Bayesian network model for establishing the relationship between traffic and population

Existing traffic prediction methods are based on previously collected traffic patterns, and the measured data are used to train and create a model to predict future traffic patterns. However, complex spatio-temporal patterns of user demand render prediction of mobile traffic, which accounts for the majority of current network, challenging. In this study, a network traffic prediction method based on a Bayesian network was proposed to model the relationship among network traffic, edge/cloud computing resource usage, and population in a target area. Although an accurate estimate of population in a target area can be obtained using conventional methods, the number of active users and the traffic and the edge/cloud computing resource usage in the target area cannot be accurately estimated. The objective of this study was to reduce the gap between population estimates and the number of active mobile users in an area. The Bayesian network parameters were estimated from the measurement of the past network traffic, edge/cloud computing resource usage, and population. In this study, we assumed that the ratio of the number of communicating users dynamically changes and presented a detailed performance evaluation using the Toy and Milan grid models. The basic performance of the proposed Bayesian network-based prediction method was investigated using a Toy model and confirmed that the proposed model achieved accurate prediction even in a scenario in which the activity factors changed. We evaluated the performance of the proposed model using the Milan grid dataset and confirmed that it achieved accurate results for both single and multiple traffic classes scenario.