Neural Network For Risk Assessment In Life Insurance Industry: A Case Study

This paper presents a system to predict the risk of non-payment of premium in health insurance. The data corpus includes a total of 186 instances divided into 127 samples (70%) for the learning phase and 59 samples (30%) for the validation and test phase. Each example is characterized by age, marital status, the presence of a recent illness or not, the wearing of medical glasses or prostheses, the gender, the recovery rate and the ceiling exceeded. After normalizing the data, an analysis has been performed to ensure non-redundancy by calculating the covariance. The error back propagation algorithm is used for the learning phase. The minimization of the quadratic error has allowed to retain the number of neurons on the hidden layer. Neuroph library is applied for the implementation. The performance of the system is rated at 88.71%.