Multidimensional Periodic Hawkes Processes for Credit Card Data
Patterns of human activity are crucial to applications ranging from resource allocation, disease spread and infrastructure development. As can be seen from mobility traces, credit card transactions, communications data and other records of activity, this kind of deliberate human behavior is inherently non-Poissonian and tends to exhibit bursts of activity throughout time. We propose a self-exciting Hawkes process model that captures circadian rhythms in activity. We show that in addition to outperforming several baseline approaches, this model can uncover the influence structure of events, lending itself to interpretable lifestyles and providing insight on collective human behavior.
Author: Sharon Xu
