A Framework for Improving Attack Detection Accuracy using Ensemble Methods

Abstract

With the development of internet and communication cyber movement started into new era. Today internet is used by all the organizations and people and they share a lot of sensitive information. A lot of attacks occur through the internet. These attacks exploit the vulnerabilities present in the system and may destroy the sensitive information present in the system. Protection mechanisms need to be provided to protect against these attacks. Firewalls and other basic security measures have been implemented to counter these attacks but these have failed as everyday novel ideas are developed by attackers to attack the system. Thus there is need to develop a system to eradicate these attacks as they damage the confidential information of the organizations. Intrusion detection systems are used for this purpose. Data mining can be used in case of intrusion detection system to differentiate between legitimate and illegitimate connections. Various classification algorithms can be applied that classify the connection either as normal or of specific attack type. Ensemble learning techniques are being currently considered as a new way to detect intrusive activities in systems as they have higher accuracy. The proposed approach is fusion of classification with boosting algorithms. In ensemble learning fusion of two or more techniques is done and accuracy of the combined system is large as compared to the individual techniques. The proposed model is applied on KDDCUP’99 dataset i.e. widely available dataset for intrusion detection systems. The results of the individual classification algorithms are compared with ensembling results. The ensemble learning better classifies the results to their proper category in terms of accuracy and number or instances properly classified.