Speaker: Christos Mitas
We harness the flexibility and extensibility of the Bayesian Hierarchical Modelling (BHM) paradigm to understand the frequency and severity of cyber incidents relating to breach of privacy and data exfiltration. As we continuously enhance our proprietary database, we collect and assess such incidents at scale. Incident characteristics include victim demographics, measures of damage severity, time stamps, and occasionally features of threat actors. These characteristics constitute clear hierarchies which can be optimally modelled by BHMs which separately represent event occurrence frequencies and their conditional severities. Additionally, since under-reporting of cyber incidents is a well-known aspect of this type of data, we build an augmentation model which represents this effect by a joint model of the data generating process and the imperfect reporting process.