Exploiting tactics, techniques, and procedures for malware detection
There has been a meteoric rise in the use of malware to perpetrate cybercrime and more generally, serve the interests of malicious actors. As a result, malware has evolved both in terms of its sheer variety and sophistication. There is hence a need for developing effective malware detection systems...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
2023
|
| Subjects: |
| _version_ | 1811139557303779328 |
|---|---|
| author | Sharma, Y |
| author2 | Martinovic, I |
| author_facet | Martinovic, I Sharma, Y |
| author_sort | Sharma, Y |
| collection | OXFORD |
| description | There has been a meteoric rise in the use of malware to perpetrate cybercrime and more generally, serve the interests of malicious actors. As a result, malware has evolved both in terms of its sheer variety and sophistication. There is hence a need for developing effective malware detection systems to counter this surge. Typically, most such systems nowadays are purely data-driven - they utilise Machine Learning (ML) based approaches which rely on large volumes of data, to spot patterns, detect anomalies, and thus detect malware. In this thesis, we propose a methodology for malware detection on networks that combines human domain knowledge with conventional malware detection approaches to more effectively identify, reason about, and be resilient to malware. Specifically, we use domain knowledge in the form of the Tactics, Techniques, and Procedures (TTPs) described in the MITRE ATT\&CK ontology of adversarial behaviour to build Network Intrusion Detection Systems (NIDS). Through the course of our research, we design and evaluate the first such NIDS that can effectively exploit TTPs for the purpose of malware detection. We then attempt to expand the scope of usability of these TTPs to systems other than our specialised NIDS, and develop a methodology that lets any generic ML-based NIDS exploit these TTPs as model features. We further expand and generalise our approach by modelling it as a multi-label classification problem, which enables us to: (i) detect malware more precisely on the basis of individual TTPs, and (ii) identify the malicious usage of uncommon or rarely-used TTPs. Throughout all our experiments, we rigorously evaluate all our systems on several metrics using large datasets of real-world malware and benign samples. We empirically demonstrate the usefulness of TTPs in the malware detection process, the benefits of a TTP-based approach in reasoning about malware and responding to various challenging conditions, and the overall robustness of our systems to adversarial attack. As a consequence, we establish and improve the state-of-the-art when it comes to detecting network-based malware using TTP-based information. This thesis overall represents a step forward in building automated systems that combine purely-data driven approaches with human expertise in the field of malware analysis. |
| first_indexed | 2024-09-25T04:07:59Z |
| format | Thesis |
| id | oxford-uuid:d8eb8401-90e4-4049-98fd-c423c47557f9 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-09-25T04:07:59Z |
| publishDate | 2023 |
| record_format | dspace |
| spelling | oxford-uuid:d8eb8401-90e4-4049-98fd-c423c47557f92024-06-03T14:41:46ZExploiting tactics, techniques, and procedures for malware detectionThesishttp://purl.org/coar/resource_type/c_db06uuid:d8eb8401-90e4-4049-98fd-c423c47557f9Intrusion detection systems (Computer security)Computer securityCyber intelligence (Computer security)Artificial intelligenceMalware (Computer software)EnglishHyrax Deposit2023Sharma, YMartinovic, IThere has been a meteoric rise in the use of malware to perpetrate cybercrime and more generally, serve the interests of malicious actors. As a result, malware has evolved both in terms of its sheer variety and sophistication. There is hence a need for developing effective malware detection systems to counter this surge. Typically, most such systems nowadays are purely data-driven - they utilise Machine Learning (ML) based approaches which rely on large volumes of data, to spot patterns, detect anomalies, and thus detect malware. In this thesis, we propose a methodology for malware detection on networks that combines human domain knowledge with conventional malware detection approaches to more effectively identify, reason about, and be resilient to malware. Specifically, we use domain knowledge in the form of the Tactics, Techniques, and Procedures (TTPs) described in the MITRE ATT\&CK ontology of adversarial behaviour to build Network Intrusion Detection Systems (NIDS). Through the course of our research, we design and evaluate the first such NIDS that can effectively exploit TTPs for the purpose of malware detection. We then attempt to expand the scope of usability of these TTPs to systems other than our specialised NIDS, and develop a methodology that lets any generic ML-based NIDS exploit these TTPs as model features. We further expand and generalise our approach by modelling it as a multi-label classification problem, which enables us to: (i) detect malware more precisely on the basis of individual TTPs, and (ii) identify the malicious usage of uncommon or rarely-used TTPs. Throughout all our experiments, we rigorously evaluate all our systems on several metrics using large datasets of real-world malware and benign samples. We empirically demonstrate the usefulness of TTPs in the malware detection process, the benefits of a TTP-based approach in reasoning about malware and responding to various challenging conditions, and the overall robustness of our systems to adversarial attack. As a consequence, we establish and improve the state-of-the-art when it comes to detecting network-based malware using TTP-based information. This thesis overall represents a step forward in building automated systems that combine purely-data driven approaches with human expertise in the field of malware analysis. |
| spellingShingle | Intrusion detection systems (Computer security) Computer security Cyber intelligence (Computer security) Artificial intelligence Malware (Computer software) Sharma, Y Exploiting tactics, techniques, and procedures for malware detection |
| title | Exploiting tactics, techniques, and procedures for malware detection |
| title_full | Exploiting tactics, techniques, and procedures for malware detection |
| title_fullStr | Exploiting tactics, techniques, and procedures for malware detection |
| title_full_unstemmed | Exploiting tactics, techniques, and procedures for malware detection |
| title_short | Exploiting tactics, techniques, and procedures for malware detection |
| title_sort | exploiting tactics techniques and procedures for malware detection |
| topic | Intrusion detection systems (Computer security) Computer security Cyber intelligence (Computer security) Artificial intelligence Malware (Computer software) |
| work_keys_str_mv | AT sharmay exploitingtacticstechniquesandproceduresformalwaredetection |