Typhoon Economic Loss Prediction in China by Applying General Regression Neural Network and Hierarchical Cluster Analysis

抄録

Natural disasters may cause extreme damage and enormous economic loss. It is important to look for efficient and precise damage prediction models using neural networks, which are increasingly used in many applications. One challenge of developing such a damage prediction model is its limited amount of available data. We therefore chose to predict typhoon damage loss based on a general regression neural network (GRNN). The GRNN is able to converge to kernel functions of data with limited training samples available. This paper investigates a GRNN-based neural network and introduces a loss prediction index. The proposed GRNN structure gives an improved prediction performance with a normalized mean squared error of 0.0071 and a correlation of 0.9321. According to prediction results of economic loss, 30 typhoons have been grouped into five categories by hierarchical cluster analysis. Due to its simplicity and fast-converging features, this scheme is suitable for practical, simple but robust typhoon damage prediction.