With the development of the Internet, many businesses will attract users by marketing activities on terminal devices, such as: Android Phone and IOS Phone. These marketing activities also attract a lot of malicious users, who usually profit from malicious behaviors. This kind of malicious behavior will cause the merchants to invest a lot of money in marketing activities without really attracting users and causing a lot of losses. Tversky trust was evaluate from the cloud nodes in order to identify nature of infected and un-infected nodes. For prediction of model virtual nodes other features were also used like memory, processor, utilization, etc. Training of error back propagation neural network was done by use of features and predict the nature of node. Experiment was done on different environment by varying number of nodes and result shows that proposed model has increases the work detection accuracy as compared to previous existing models.