Combating Cyber Security Breaches in Digital World Using Misuse Detection Methods: Misuse Detection

Combating Cyber Security Breaches in Digital World Using Misuse Detection Methods: Misuse Detection

Subbulakshmi T.
Copyright: © 2016 |Pages: 8
DOI: 10.4018/978-1-5225-0193-0.ch006
OnDemand:
(Individual Chapters)
Available
$37.50
Current Special Offers
No Current Special Offers
TOTAL SAVINGS: $37.50

Abstract

Intrusion Detection Systems (IDS) play a major role in the area of combating security breaches for information security. Current IDS are developed with Machine learning techniques like Artificial Neural Networks, C 4.5, KNN, Naïve Bayes classifiers, Genetic algorithms Fuzzy logic and SVMs. The objective of this paper is to apply Artificial Neural Networks and Support Vector Machines for intrusion detection. Artificial Neural Networks are applied along with faster training methods like variable learning rate and scaled conjugate gradient. Support Vector Machines use various kernel functions to improve the performance. From the kddcup'99 dataset 45,657 instances are taken and used in our experiment. The speed is compared for various training functions. The performance of various kernel functions is assessed. The detection rate of Support Vector Machines is found to be greater than Artificial Neural Networks with less number of false positives and with less time of detection.
Chapter Preview
Top

2. Background

Soft computing was first proposed by Lotfi Zadeh to construct new generation computationally intelligent hybrid intelligent systems including Neural networks, Fuzzy logic, approximate reasoning and derivative free optimization techniques like Genetic algorithms for most of the real world applications like function approximation, Image processing and Intrusion detection.

Neural Networks are applied to effectively find out the intrusions by Martin Botha, Rossouw von solms. Next Generation Proactive Identification Model is used to protect the system using neural networks. This model is based on the assumption that each user's behaviour is unique and when he leaves the system it could be recorded for further comparison. Fuzzy Default Logic(FDL) (Jian et al., 2003)is applied to Intrusion Detection using reasoning and FDL-IDS was developed which increases detection speed and accuracy which reduces the cumulative cost of developing an traditional Intrusion Detection Expert System (Denning, 1987). A Novel attack detection method is proposed to detect intrusions using fuzzy logic and data mining. The proposed system is designed as both misuse and anomaly detection system by combining well-formed fuzzy if-then rules and simple data mining.

ANN and SVM are employed for intrusion detection using (Mukkamala & Sung, 2005) DARPA dataset and it is observed that SVMs are superior to ANN in three critical respects. SVM train and run an order of magnitude faster; SVM scale much faster; and SVMs give high classification accuracy. Feature selection and ranking is presented using two methods: first is Performance Based Ranking Method using performance metrics and 34 out of 41 features are selected for classification by a Multilayer FF ANN. Second using SVMs with support vector decision function. 23 out of 41 features are selected in this method for classification. SVMs prove higher classification accuracy.

Complete Chapter List

Search this Book:
Reset