A hybrid machine learning/statistical model of Grid security

Most current Grid security techniques concentrate on traditional security aspects such as authentication, authorization, etc. While they have shown their usefulness, the significance of the information hidden in the historical data corpus denoting the user-Grid interactions has been largely neglected. In fact, such information provide great insight into Grid security and if properly harnessed, will help better protect the Grid against potential attacks. To utilize these hidden information in a service-oriented Grid environment, we propose a hybrid machine learning and statistical model. The machine learning component predicts the security of a service by considering the probability distribution of the past services, while the statistical component evaluates a service's security statistically based on its own past behaviors and users' opinions. We construct an overall architecture based on this hybrid model and demonstrate through examples its effectiveness and potential to offer stronger security to the Grid.