DESIGN OF A NEURAL-NETWORK FOR RECOGNITION AND CLASSIFICATION OF COMPUTER VIRUSES

A sufficient number of experiments has been conducted to ascertain that a neural network can be used as a component of a computer security system for the recognition and classification of computer virus attack A set of attributes that describe the system activity and the behaviour of computer viruses has been identified. The error back propagation training algorithm and the self-organizing feature map have been studied. Several experiments were conducted using both algorithms, different learning parameters, and two different training sets. For each architecture, the size of the network with the best performance was estimated experimentally. Results indicate that both neural networks can discriminate input patterns, at almost the same level of accuracy. The number of neurons required for the solution of the specific problem using a multilayer perceptron network was smaller than the respective number for a self-organizing feature map network Therefore, using back propagation, the training and the recall process were faster. In conclusion, neural networks were proved to be efficient and practical devices for computer virus recognition and classification, in certain environments.