Chapter 10: Multivariate Copulas Model in Spatiotemporal Irregular Pattern Detection in Mobility Network

Characterizing localized mobility network traffic is one of the most challenging tasks in mobility network planning. Network dimensioning does not only need to consider statistically steady-state traffic, but it also needs to take into account the situations when special events happen in some areas, especially when the events incur intensive traffic loads even though the occurrence is rare. This chapter proposes a multivariate copulas concept to identify the areas that have different traffic patterns compared with their neighbors. Multivariate dependence statistic pseudo is constructed to measure the spatial relationship among the multiple neighbor time series, which is based on the degree of dependence for multivariate extreme value copula. Weighted dependency, which integrates temporal extreme value features with spatial dependency structure, is established to detect the areas that have irregular temporal traffic patterns. A new spatial neighbor detection procedure is illustrated to obtain the areas that are robust to irregular shapes. A synthetic dataset that simulates the real network traffic is generated to illustrate our procedure and validate the performance.