A Study On The Spatial-time Early Warning Models Of Infectious Disease Based On Bayesian Network

With the constantly appearance of a variety of new infectious diseases and the renewed prevalence of existing diseases, to establish the early warning system for these diseases is particularly important. In recent years, countries around the world have conducted extensive research on assessing the probabilities of infectious disease outbreaks in different dimensions and different scales mainly including time or spatial dimension, few can simultaneously contain both the spatial and time characteristics. Moreover, traditional methods estimate the risks of infectious diseases mainly based on anomalies that are detected in a disease’s distribution, which entails less timely retrospective analyses, but methods based on analysing the impacts of risk factors are overly dependent on the acquisition of risk data. In this article, we firstly introduced some space aggregation detection technologies of infectious diseases, especially the cluster analysis of spatial panel data; secondly we discussed some time aggregation detection technologies such as moving percentile method; at last, we proposed a new method using a Bayesian belief network(BBN) to predict the risk levels of infectious diseases; this method can make uncertainty estimates even with missing values. As a complement, we applied these methods to a common infectious disease of hand, foot and mouth disease in Hunan province, China. In this study, we divided the study area into four parts based on the cluster algorithm of multiple indicator panel data, and established different BBN models for different areas, the accuracy of final results can reach from 86.11% to 100%. The experimental results show not only that this method based on a BBN has higher accuracy than traditional methods but also that once the relational network is built between disease risk and the risk factors, we can also assess the risk even when some data are missing.