Thesis

Using Neural Networks in Intrusion Detection System for Cloud Computing

Information security is an issue of global concern nowadays. The complexity, accessibility, and extensiveness of the Internet have led to tremendous increase of security risk of information systems. The situation hasn't been improved and the harm and loss an intruder or a malicious attack can cause to an information system are known by more and more people, especially when the term "cloud computing" appeared in people's sights. 
 
 Because of the grid distribution of cloud computing users, and their lack of knowledge on managing the cloud services, these users and their systems are always regarded as easy targets for intruders looking for possible vulnerabilities. In this case, an intrusion detection system (IDS) which collects user behaviors in the network and detects malicious activities can protect our systems from the attacks. Neural networks are widely applied in designing IDSs, because of their adaptation and qualifications in dealing with large scale of data, in our case, the dramatically large quantities of user behaviors through the Internet.
 
 Recently, in the area of artificial neural networks, the concept of combining multiple networks has been proposed as a fresh direction for the design and development of highly reliable IDS.
 
 An intrusion detection system built based on Multilayer perceptron (MLP) neural network and k-means neural network is presented in this thesis; the system combines both supervised learning and unsupervised learning methods and outperforms the intrusion detecting accuracy rate from systems based on either of them. The key idea of this system is to discover useful patterns or features that describe user behavior on a system, and this set of relevant features is used to build classifiers that can distinguish anomalies and known intrusions with normal user behaviors. Using a set of benchmark data from a KDD (Knowledge Discovery and Data Mining) competition supported by DARPA (Defense Advanced Research Projects Agency), the efficiency and accuracy this proposed system can achieve is demonstrated and the comparison between the performances of proposed system and other IDSs using only MLP, K-means and other neural networks or techniques like SOM (Self-organized Maps) and Radial Basis Function (RBF) are presented.
 
 
 Keywords: Cloud Computing, Intrusion Detection System (IDS), Artificial Neural Network (ANN), Supervised Learning, Unsupervised Learning, Multilayer Perceptron (MLP), K-means Algorithm

Information security is an issue of global concern nowadays. The complexity, accessibility, and extensiveness of the Internet have led to tremendous increase of security risk of information systems. The situation hasn't been improved and the harm and loss an intruder or a malicious attack can cause to an information system are known by more and more people, especially when the term "cloud computing" appeared in people's sights. Because of the grid distribution of cloud computing users, and their lack of knowledge on managing the cloud services, these users and their systems are always regarded as easy targets for intruders looking for possible vulnerabilities. In this case, an intrusion detection system (IDS) which collects user behaviors in the network and detects malicious activities can protect our systems from the attacks. Neural networks are widely applied in designing IDSs, because of their adaptation and qualifications in dealing with large scale of data, in our case, the dramatically large quantities of user behaviors through the Internet. Recently, in the area of artificial neural networks, the concept of combining multiple networks has been proposed as a fresh direction for the design and development of highly reliable IDS. An intrusion detection system built based on Multilayer perceptron (MLP) neural network and k-means neural network is presented in this thesis; the system combines both supervised learning and unsupervised learning methods and outperforms the intrusion detecting accuracy rate from systems based on either of them. The key idea of this system is to discover useful patterns or features that describe user behavior on a system, and this set of relevant features is used to build classifiers that can distinguish anomalies and known intrusions with normal user behaviors. Using a set of benchmark data from a KDD (Knowledge Discovery and Data Mining) competition supported by DARPA (Defense Advanced Research Projects Agency), the efficiency and accuracy this proposed system can achieve is demonstrated and the comparison between the performances of proposed system and other IDSs using only MLP, K-means and other neural networks or techniques like SOM (Self-organized Maps) and Radial Basis Function (RBF) are presented. Keywords: Cloud Computing, Intrusion Detection System (IDS), Artificial Neural Network (ANN), Supervised Learning, Unsupervised Learning, Multilayer Perceptron (MLP), K-means Algorithm

Relationships

Items