Publication

Abstract : Modeling of claim frequency is a vital factor in the non-life insurance industry. The claim count data in non-life insurance may not follow the traditional regression count data models with the use of Poisson or negative binomial distribution in numerous circumstances due to excessive number of zeros in the real data set. If the excessive number of zeros is not considered with sufficient weight, it will lead to information shortage to get a accurate rate making for the non-life insurance portfolio. In this paper we compared different claim count models such as zero-inflated Poisson (ZIP) regression model, hurdle model with back propagation neural network (BPNN) for modeling the count data which has excessive number of zeros. We shown from our empirical study that BPNN outperforms the conventionally used models and provided better fit to claim count data in terms of mean squared error (MSE).