There are two ways to model a low probability event. a) consider the low probability event in the presence of other common events and treat the event as an anomaly or novelty and train novelty detection algorithms on it, and b) Track relevant variables before and after the low probability event occurs to detect causality