Malware attack is becoming one of the most threats to internet security. Botnet in specific, used to generate spam, carry out DDOS attacks, steal sensitive information becoming major threats for committing cybercrimes. In this paper, we propose an artificial neural network model to predict types of botnet for the next day attack. In this experiment, several number of hidden neurons are manipulated in order to minimize error. Moreover, in order to minimize the processing time, the model is running on graphical processing unit (GPU) and the performance is compared to computational processing unit (CPU). The experimental results indicate that the model produce lowest error on 500 number of hidden neurons and there are significant different between the running time between GPU and CPU.