Phillip Yam (Chinese University of Hong Kong)
Classifying severity of risks has long been of vital importance in insurance and finance, this is a major concern in InsurTech and FinTech. Among widely adopted classifiers in practical use, the application of Support Vector Machine, Neural Network and Logistic Regression to insurance and finance datasets would lead to a potential substantial loss of information as these datasets usually involve a lot of categorical variables, yet none of these classifiers handle them comprehensively; 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 deal with continuous feature variables. Moreover, the relatively strong dependence structures among feature variables, especially among categorical feature variables, in insurance and finance practices have not been explicitly accounted for in the aforementioned existing classifiers. We here propose to effectively model such an implicit strong enough dependence by comonotonicity. Altogether will be dealt with through our newly proposed Comonotone-Independence Bayes Classifier (CIBer), this leads to a far better clustering of the predictive feature variables that can facilitate an effective classification. We shall also demonstrate the effectiveness of CIBer as a tool in data analytics against those common classifiers through the empirical studies upon several representative datasets in finance and insurance.