Abstract
Predicting frequencies and classifying severities of risks has long been of vital importance, this is a major concern in InsurTech, FinTech, and the emerging field of cyber risk security. Among widely adopted classifiers in practical use, the application of Support Vector Machine, Neural Network (e.g. MLP), Generalized Linear Models and Linear Discriminant Analysis to insurance and finance datasets would lead to a potential substantial loss of information as these datasets usually involve a lot of categorical feature, yet none of these classifiers handle them comprehensively; this issue is even more alerting in cyber risk datasets, with a majority of features as categorical, including but not limited to the types of entity, breach and industry. On the other hand, Classification and Regression Tree handles categorical and discrete feature variables well enough by its design, yet it lacks the mechanism to cope with continuous features.
Moreover, the relatively strong dependence structures among feature variables, especially among categorical features, in insurance and financial practices have not been explicitly accounted for in these prevalent classifiers.
In this talk, we propose to effectively model an implicit strong enough dependence by comonotonicity, and perform risk classification through our newly proposed Comonotone-Independence Bayes Classifier (CIBer), this leads to a far better clustering of the predictive feature variables. As a result, CIBer can facilitate an effective classification, and it can also be enhanced to serve as a powerful regressor against all types of feature variables. We shall demonstrate the effectiveness and resulting profitability of CIBer as a tool in data analytics against all others through the empirical studies upon several representative datasets in finance and insurance. Besides, due to the time, we just sketch out its immediate use in cyber risk detection through the enhanced modelling of superposed marked Hawkes processes; indeed, it gives an impressive improvement in the prediction power of both claim frequency and severity.
CIBer won a Silver Medal in the 48th International Exhibition of Inventions Geneva in 2023. The Python code can be found at https://github.com/kaiser1999/CIBer.
Biography
Prof. Phillip Yam received his BSc in Actuarial Science, with First Class Honours and Dean's Listings, and MPhil under supervision of Professor Hailiang Yang from the University of Hong Kong. Supported by the two scholarships awardedby the Croucher Foundation, he obtained an MASt (Master of Advanced Study) degree, Part III of the Mathematical Tripos, with Distinction in Mathematics from the University of Cambridge, and a DPhil in Mathematics under supervision of Professor Terry Lyons from the University of Oxford. During his postgraduate studies, he was also awarded the E. M.Burnett Prize in Mathematics from the University of Cambridge, and the junior research fellowship from The Erwin Schrödinger International Institute for Mathematics and Physics of the University of Vienna. Phillip is currently Director of the Quantitative Finance and Risk Management Science Programme, and a Professor at the Department of Statistics and Data Science of the Chinese University of Hong Kong (CUHK); he is also Assistant Dean (Education) of CUHK Faculty of Science. He was appointed as a research fellow in the Hausdorff Research Institute for Mathematics of the University of Bonn in Germany, a Visiting Professor in both the Department of Statistics of Columbia University in the City of New York and the Naveen Jindal School at the University of Texas at Dallas in the United States of America, as well as a Distinguished Visiting Scholar at the School of Risk and Actuarial Studies in the University of New South Wales in Australia. In June 2026, he became an Elected Member of the International Statistical Institute.
He has published more than a hundred journal articles in actuarial science, applied mathematics, control theory and engineering, data analytics, financial mathematics and economics, operations management, probability and stochastic analysis, and statistics. He is currently an Editor of Insurance: Mathematics and Economics, a top journal in actuarial science, and also serves on editorial boards of several journals. Besides, he wrote the first ever monograph on mean field theory, Mean Field Games and Mean Field Type Control Theory, and another one called Financial Data Analytics with Machine Learning, Optimization and Statistics; one of his original research outputs included in the second book, “Comonotone-independence Bayes Classifier (CIBer),” was also awarded a Silver Medal in the 48th International Exhibition of Inventions Geneva in 2023. Besides, he has served as an external examiner for various funding institutions and universities in Hong Kong, Europe, North America, and Asia-Pacific regions, and a panel member of the selection committee for the Croucher Study Award. He has supervised over 30 postgraduate students and postdocs, many of them are now outstanding faculty members in world-renowned institutions, while others are expert practitioners such as quants and iBankers in financial and insurance industries, as well as international data analytics corporations.
