Risk Analysis Model Based on Prototype Theory

In risk analysis in health insurance, one fundamental duty is to classify the insured into the risk categories which are essentially the vague concepts. A defining feature of vague concepts that their boundaries are uncertain. Inthis paper, we take a prototype theory approach to represent risk categories. In this representation, each label is associated with a prototype and has a probability density function. One main contribution of this paper is that a learning procedure is proposed to obtain a label representation from the given data set. The idea of learning algorithm is that the training dataset can be considered as the samples generated from the label, and hence the data set has a generating probability distribution from the underlying label. By taking the Mahalanobis distance to measure the distance between elements and label, an iterative learning algorithm is proposed to obtain the label representation.The basic procedure includes two iterations. The first is to generate a label representation from the training data set with a probability distribution. The second is to generate the probability distribution associated with the training data from the label representation. These two iterations continue till the termination condition is satisfied.