Bayesian Estimation Of Security Risks Distribution And Prediction Application

Risk is a hot topic of current research. How to define a security risk metrics to reflect thedifferences in sampling information is an urgent problem in practical application. Further, wecan predict and control the risk according to the distribution of related factors.This paper proposes a new definition of security risks appropriating the risk of theresource allocation problem. A risk assessment and prediction model has been built based onthe research of the distribution of difference information. The model is applied well to predictthe risk of resource allocation of university.The main research of this paper is as follows:(1) The security risk is defined as the uncertainty of the deviation caused by sampling andtesting information from various factors, and considered security risks are due to the effect ofdifferences in values and differences possibility. Based on the performance evaluation model,the performance difference information is defined to the difference information causing riskof resource allocation of university.(2) Bayesian method is used to estimate the parameters of distribution of differenceinformation dynamically thereby improving the reliability of risk prediction. In the allocationrisk, a hierarchical bayes model is established to dynamically estimate the parameters of thenormal distribution of performance difference information. The results show that thedistribution of performance difference information fluctuates less over the years.(3) Seeking centers of different risk levels from data using K-Means clustering algorithmand using Gaussian membership function to describe the risk fuzzily. Then the riskassessment model has been built and applied to the allocation risk problem in good agreementwith the current situation of university.(4) Based on the results of the risk assessment, the risk prediction model is established topredicte the range of comprehensive risk level based on BP neural network. In the allocationrisk problem the prediction accuracy rate reaches93.06%.