<p dir="ltr">The explosive growth of computer networks over the past few decades has significantly enhanced communication capabilities. However, this expansion has also attracted malicious attackers seeking to compromise and disable these networks for personal gain. Network Intrusion Detection Systems (NIDS) were developed to detect threats and alert users to potential attacks. As the types and methods of attacks have grown exponentially, NIDS have struggled to keep pace. A paradigm shift occurred when NIDS began using Machine Learning (ML) to differentiate between anomalous and normal traffic, alleviating the challenge of tracking and defending against new attacks. However, the adoption of ML-based anomaly detection in NIDS has unraveled a new avenue of exploitation due to the inherent inadequacy of machine learning models - their susceptibility to adversarial attacks.</p><p dir="ltr">In this work, we explore the application of adversarial attacks from the image domain to bypass Network Intrusion Detection Systems (NIDS). We evaluate both white-box and black-box adversarial attacks against nine popular ML-based NIDS models. Specifically, we investigate Projected Gradient Descent (PGD) attacks on two ML models, transfer attacks using adversarial examples generated by the PGD attack, the score-based Zeroth Order Optimization attack, and two boundary-based attacks, namely the Boundary and HopSkipJump attacks. Through comprehensive experiments using the NSL-KDD dataset, we find that logistic regression and multilayer perceptron models are highly vulnerable to all studied attacks, whereas decision trees, random forests, and XGBoost are moderately vulnerable to transfer attacks or PGD-assisted transfer attacks with approximately 60 to 70% attack success rate (ASR), but highly susceptible to targeted HopSkipJump or Boundary attacks with close to a 100% ASR. Moreover, SVM-linear is highly vulnerable to both transfer attacks and targeted HopSkipJump or Boundary attacks achieving around 100% ASR, whereas SVM-rbf is highly vulnerable to transfer attacks with a 77% ASR but only moderately to targeted HopSkipJump or Boundary attacks with a 52% ASR. Finally, both KNN and Label Spreading models exhibit robustness against transfer-based attacks with less than 30% ASR but are highly vulnerable to targeted HopSkipJump or Boundary attacks with a 100% ASR with a large perturbation. Our findings may provide insights for designing future NIDS that are robust against potential adversarial attacks.</p>

The tremendous increase in usage and complexity of modern communication and network systems connected to the Internet, places demands upon security management to protect organisations' sensitive data and resources from malicious intrusion. Malicious attacks by intruders and hackers exploit flaws and weakness points in deployed systems through several sophisticated techniques that cannot be prevented by traditional measures, such as user authentication, access controls and firewalls. Consequently, automated detection and timely response systems are urgently needed to detect abnormal activities by monitoring network traffic and system events. Network Intrusion Detection Systems (NIDS) and Network Intrusion Prevention Systems (NIPS) are technologies that inspect traffic and diagnose system behaviour to provide improved attack protection. The current implementation of intrusion detection systems (commercial and open-source) lacks the scalability to support the massive increase in network speed, the emergence of new protocols and services. Multi-giga networks have become a standard installation posing the NIDS to be susceptible to resource exhaustion attacks. The research focuses on two distinct problems for the NIDS: missing alerts due to packet loss as a result of NIDS performance limitations; and the huge volumes of generated alerts by the NIDS overwhelming the security analyst which makes event observation tedious. A methodology for analysing alerts using a proposed framework for alert correlation has been presented to provide the security operator with a global view of the security perspective. Missed alerts are recovered implicitly using a contextual technique to detect multi-stage attack scenarios. This is based on the assumption that the most serious intrusions consist of relevant steps that temporally ordered. The pre- and post- condition approach is used to identify the logical relations among low level alerts. The alerts are aggregated, verified using vulnerability modelling, and correlated to construct multi-stage attacks. A number of algorithms have been proposed in this research to support the functionality of our framework including: alert correlation, alert aggregation and graph reduction. These algorithms have been implemented in a tool called Multi-stage Attack Recognition System (MARS) consisting of a collection of integrated components. The system has been evaluated using a series of experiments and using different data sets i.e. publicly available datasets and data sets collected using real-life experiments. The results show that our approach can effectively detect multi-stage attacks. The false positive rates are reduced due to implementation of the vulnerability and target host information.

Submitted by VINÍCIUS OLIVEIRA FERREIRA null (viniciusoliveira@acmesecurity.org) on 2016-05-18T20:29:41Z No. of bitstreams: 1 Dissertação-mestrado-vinicius-oliveira-biblioteca-final.pdf: 1594758 bytes, checksum: 0dbb0d2dd3fca3ed2b402b19b73006e7 (MD5) / Approved for entry into archive by Ana Paula Grisoto (grisotoana@reitoria.unesp.br) on 2016-05-20T16:27:30Z (GMT) No. of bitstreams: 1 ferreira_vo_me_sjrp.pdf: 1594758 bytes, checksum: 0dbb0d2dd3fca3ed2b402b19b73006e7 (MD5) / Made available in DSpace on 2016-05-20T16:27:30Z (GMT). No. of bitstreams: 1 ferreira_vo_me_sjrp.pdf: 1594758 bytes, checksum: 0dbb0d2dd3fca3ed2b402b19b73006e7 (MD5) Previous issue date: 2016-04-27 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Os Sistemas de Detecção de Intrusão baseados em rede (NIDS) são tradicionalmente divididos em dois tipos de acordo com os métodos de detecção que empregam, a saber: (i) detecção por abuso e (ii) detecção por anomalia. Aqueles que funcionam a partir da detecção de anomalias têm como principal vantagem a capacidade de detectar novos ataques, no entanto, é possível elencar algumas dificuldades com o uso desta metodologia. Na detecção por anomalia, a análise das anomalias detectadas pode se tornar dispendiosa, uma vez que estas geralmente não apresentam informações claras sobre os eventos maliciosos que representam; ainda, NIDSs que se utilizam desta metodologia sofrem com a detecção de altas taxas de falsos positivos. Neste contexto, este trabalho apresenta um modelo para a classificação automatizada das anomalias detectadas por um NIDS. O principal objetivo é a classificação das anomalias detectadas em classes conhecidas de ataques. Com essa classificação pretende-se, além da clara identificação das anomalias, a identificação dos falsos positivos detectados erroneamente pelos NIDSs. Portanto, ao abordar os principais problemas envolvendo a detecção por anomalias, espera-se equipar os analistas de segurança com melhores recursos para suas análises. / Network Intrusion Detection Systems (NIDS) are traditionally divided into two types according to the detection methods they employ, namely (i) misuse detection and (ii) anomaly detection. The main advantage in anomaly detection is its ability to detect new attacks. However, this methodology has some downsides. In anomaly detection, the analysis of the detected anomalies is expensive, since they often have no clear information about the malicious events they represent; also, it suffers with high amounts of false positives detected. In this context, this work presents a model for automated classification of anomalies detected by an anomaly based NIDS. Our main goal is the classification of the detected anomalies in well-known classes of attacks. By these means, we intend the clear identification of anomalies as well as the identification of false positives erroneously detected by NIDSs. Therefore, by addressing the key issues surrounding anomaly based detection, our main goal is to equip security analysts with best resources for their analyses.

Security is one of the major issues in any network and on the Internet. It encapsulates many different areas, such as protecting individual users against intruders, protecting corporate systems against damage, and protecting data from intrusion. It is obviously impossible to make a network totally secure, as there are so many areas that must be protected. This thesis includes an evaluation of current techniques for internal misuse of computer systems, and tries to propose a new way of dealing with this problem. This thesis proposes that it is impossible to fully protect a computer network from intrusion, and shows how different methods are applied at differing levels of the OSI model. Most systems are now protected at the network and transport layer, with systems such as firewalls and secure sockets. A weakness, though, exists in the session layer that is responsible for user logon and their associated password. It is thus important for any highly secure system to be able to continually monitor a user, even after they have successfully logged into the system. This is because once an intruder has successfully logged into a system, they can use it as a stepping-stone to gain full access (often right up to the system administrator level). This type of login identifies another weakness of current intrusion detection systems, in that they are mainly focused on detecting external intrusion, whereas a great deal of research identifies that one of the main problems is from internal intruders, and from staff within an organisation. Fraudulent activities can often he identified by changes in user behaviour. While this type of behaviour monitoring might not be suited to most networks, it could be applied to high secure installations, such as in government, and military organisations. Computer networks are now one of the most rapidly changing and vulnerable systems, where security is now a major issue. A dynamic approach, with the capacity to deal with and adapt to abrupt changes, and be simple, will provide an effective modelling toolkit. Analysts must be able to understand how it works and be able to apply it without the aid of an expert. Such models do exist in the statistical world, and it is the purpose of this thesis to introduce them and to explain their basic notions and structure. One weakness identified is the centralisation and complex implementation of intrusion detection. The thesis proposes an agent-based approach to monitor the user behaviour of each user. It also proposes that many intrusion detection systems cannot cope with new types of intrusion. It thus applies Bayesian statistics to evaluate user behaviour, and predict the future behaviour of the user. The model developed is a unique application of Bayesian statistics, and the results show that it can improve future behaviour prediction than existing ARIMA models. The thesis argues that the accuracy of long-term forecasting questionable, especially in systems that have a rapid and often unexpected evolution and behaviour. Many of the existing models for prediction use long-term forecasting, which may not be the optimal type for intrusion detection systems. The experiments conducted have varied the number of users and the time interval used for monitoring user behaviour. These results have been compared with ARIMA, and an increased accuracy has been observed. The thesis also shows that the new model can better predict changes in user behaviour, which is a key factor in identifying intrusion detection. The thesis concludes with recommendations for future work, including how the statistical model could be improved. This includes research into changing the specification of the design vector for Bayesian. Another interesting area is the integration of standard agent communication agents, which will make the security agents more social in their approach and be able to gather information from other agents

The tremendous increase in usage and complexity of modern communication and network systems connected to the Internet, places demands upon security management to protect organisations¿ sensitive data and resources from malicious intrusion. Malicious attacks by intruders and hackers exploit flaws and weakness points in deployed systems through several sophisticated techniques that cannot be prevented by traditional measures, such as user authentication, access controls and firewalls. Consequently, automated detection and timely response systems are urgently needed to detect abnormal activities by monitoring network traffic and system events. Network Intrusion Detection Systems (NIDS) and Network Intrusion Prevention Systems (NIPS) are technologies that inspect traffic and diagnose system behaviour to provide improved attack protection. The current implementation of intrusion detection systems (commercial and open-source) lacks the scalability to support the massive increase in network speed, the emergence of new protocols and services. Multi-giga networks have become a standard installation posing the NIDS to be susceptible to resource exhaustion attacks. The research focuses on two distinct problems for the NIDS: missing alerts due to packet loss as a result of NIDS performance limitations; and the huge volumes of generated alerts by the NIDS overwhelming the security analyst which makes event observation tedious. A methodology for analysing alerts using a proposed framework for alert correlation has been presented to provide the security operator with a global view of the security perspective. Missed alerts are recovered implicitly using a contextual technique to detect multi-stage attack scenarios. This is based on the assumption that the most serious intrusions consist of relevant steps that temporally ordered. The pre- and post- condition approach is used to identify the logical relations among low level alerts. The alerts are aggregated, verified using vulnerability modelling, and correlated to construct multi-stage attacks. A number of algorithms have been proposed in this research to support the functionality of our framework including: alert correlation, alert aggregation and graph reduction. These algorithms have been implemented in a tool called Multi-stage Attack Recognition System (MARS) consisting of a collection of integrated components. The system has been evaluated using a series of experiments and using different data sets i.e. publicly available datasets and data sets collected using real-life experiments. The results show that our approach can effectively detect multi-stage attacks. The false positive rates are reduced due to implementation of the vulnerability and target host information.

The rampant growth of the Internet has been coupled with an equivalent growth in cyber crime over the Internet. With our increased reliance on the Internet for commerce, social networking, information acquisition, and information exchange, intruders have found financial, political, and military motives for their actions. Network Intrusion Detection Systems (NIDSs) intercept the traffic at an organization’s periphery and try to detect intrusion attempts. Signature-based NIDSs compare the packet to a signature database consisting of known attacks and malicious packet fingerprints. The signatures use regular expressions to model these intrusion activities. This thesis presents a memory efficient pattern matching system for the class of regular expressions appearing frequently in the NIDS signatures. Proposed Cascaded Automata Architecture is based on two stage automata. The first stage recognizes the sub-strings and character classes present in the regular expression. The second stage consumes symbol generated by the first stage upon receiving input traffic symbols. The basic idea is to utilize the research done on string matching problem for regular expression pattern matching. We formally model the class of regular expressions mostly found in NIDS signatures. The challenges involved in using string matching algorithms for regular expression matching has been presented. We introduce length-bound transitions, counter-based states, and associated counter arrays in the second stage automata to address these challenges. The system uses length information along with counter arrays to keep track of overlapped sub-strings and character class based transition. We present efficient implementation techniques for counter arrays. The evaluation of the architecture on practical expressions from Snort rule set showed compression in number of states between 50% to 85%. Because of its smaller memory footprint, our solution is suitable for both software based implementations on network chips as well as FPGA based designs.