A lightweight neural classifier for intrusion detection

Abstract EN

Intrusion detection and prevention is a set of techniques that try to detect attacks as they occur or after the attacks took place.

There are two recent and useful approaches to detect intrusions: misuse and anomaly.

They collect network trac activities from some points on the network or computer system and then use them to secure the network using one or both of the available detection methods.

The IDPS su er major vulnerabilities with large generation of false positives and negatives.

The anomaly detection aims to specify behavior detection problems that require modeling of pro le preliminary.

This paper describes a new approach of intrusion detection based on speci ed pro le built from training basis using a database that contains normal activities collected within monitored network.

The modeling of pro le represents a real challenge for network administrators and computer security researchers.

Our main goal is in the rst hand, to present an application of multilayer perceptron to make a monitored system, in the second hand, to build a classi er for trac events.

A supervised algorithm is suggested and used in training.

The recognition phase aims to validate the new classi er.

Our classi er is able to distinct between normal activity and intrusion.

We describe in details our novel detection approach and we validate the proposed solutions.

We demonstrated that this novel approach is robust, exible and gives useful performances using multilayer perceptron.