Comonotone-Independence Bayes Classifier (CIBer): A Robust and Effective FinTech and InsurTech Tool

In the era of big data, there are often a large scale of data available in the industry of Finance and Insurance. There are also immediate needs for companies to use the data to better classify the clients' risks and predict their needs. Our innovation can provide with a far better prediction results compared to the existing methods.

• Existing Bayes classifiers assume all feature variables are conditional independent. CIBer can model the variables by comonotonicity and independence, resulting in a better performance.
• We propose a novel joint encoding scheme, to encode the categorical feature variables in order to better extract their dependence structure.
• CIBer can be extended to the context of regression, traditional bayes classifier is only limited to classification.

• CIBer can handle all discrete, continous and categorical feature variables well because of its unique feature engineering.
• CIBer separates feature variables into different comontonic clusters, and the results are also interpretable while those from other classifiers are not.
• With the better modelling of feature variables, CIBer often outperforms existing machine learning methods.
• Our method can be easily implemented to many real-world dataset.

• Silver award, the 48th Geneva International Exhibition of Inventions

• The central issue in insurance and finance is customer risk classification, CIBer can be used to perform accurate classification in the presence of many categorical variables.
• CIBer can be applied to companies' financial statements and predict their future returns.
• As an explainable AI, CIBer may be preferred by companies as it generates outputs in alignment with compliance requirements.
• There are also a lot of potential applications in other fields, for example, education, psychology and marketing, while CIBer can also be used in regression tasks besides classification.

• US Provisional appn no. 63/487,282