Network based intrusion detection are the most deployed IDS. They frequently rely on signature matching detection method and focus on the security of low level network protocols. Because of the large number of false positives from one side, and the incapacity to detect some attack types from another side, IDS must allow more interest to the monitoring of application level protocols.Classical Group Testing, to extricate events that are individually positive is an efficient experimental procedure to detect Application-Layer attacks. False positive rates are reduced by using Pattern Mining approaches and Group Testing methods which in turn help identifying malicious clients and hacktivists. The architecture aims at proposing the technique to detect Application-Layer attacks using a novel algorithm called Eccentric Classifiers. © EuroJournals Publishing, Inc. 2012.