An Effective Two-Stage Neural Network Model and Its Application on Flood Loss Prediction

Abstract

In this study, a two-stage radial basis function neural network based model is employed to develop flood loss prediction model for insurance company. In the first stage, self-organizing map clustering is used to measuring the similarity of the input data, unlike k-means approach, number and centers of clusters can be determined dynamically and automatically. The Value-related Self Organizing Map (VSOM) is proposed to improve predicting accuracy for high loss subject. During the second stage, the weights from the hidden layer to output layer are determined by multivariate linear regression method. Simulation results show that the proposed approach can be applied successfully to build flood loss prediction models and provide higher accuracy and more direct decision support compared to current approaches.