Life of Research
Appearance
Journal article Open access CC BY 4.0 Version 4

Artificial Intelligence-Driven Cybersecurity: Developing Intelligent Frameworks for Advanced Threat Detection and Prevention

  • Dominic Namachila
  1. University of Cape Town

Journal article

LOR-00199

Download PDF (506.2 KB) Cite

Cite

Artificial Intelligence-Driven Cybersecurity: Developing Intelligent Frameworks for Advanced Threat Detection and Prevention

Read

Abstract

Here is the abstract as it appears in the thesis:

The accelerating scale, speed and sophistication of cyber attacks have outpaced the capabilities of signature-based and rule-driven security controls. Advanced persistent threats (APTs), polymorphic malware, zero-day exploits, and increasingly AI-assisted adversaries operate across cloud, enterprise, Internet of Things (IoT) and operational-technology environments, producing volumes of telemetry that human analysts cannot triage in time. Artificial intelligence (AI), and machine learning (ML) in particular, offers a path towards security systems that learn normal behaviour, generalise to previously unseen attacks, and respond at machine speed. However, existing AI-based security research is frequently limited by unrealistic evaluation, dataset obsolescence, high false-positive rates, poor explainability, vulnerability to adversarial manipulation, and weak integration with operational workflows.

This thesis proposes, implements and evaluates an intelligent, modular framework for advanced threat detection and prevention that addresses these limitations in an integrated manner. The framework combines (i) a multi-source data ingestion and feature-engineering layer; (ii) a hybrid detection engine that fuses supervised, unsupervised and graph-based learning to identify both known and novel threats; (iii) an adversarial-robustness and drift-management layer that hardens models against evasion and poisoning and maintains performance under concept drift; (iv) an explainability and analyst-assistance layer that produces human-interpretable evidence mapped to the MITRE ATT&CK knowledge base; and (v) an automated, policy-governed prevention and response layer that supports human-in-the-loop containment.

The framework is evaluated through a mixed-methods design comprising benchmark experiments on contemporary public datasets, cross-dataset and temporal generalisation tests, adversarial robustness assessment, ablation studies, latency and resource profiling, and expert evaluation by security practitioners. Results are reported in terms of detection performance, false-alarm rate, time-to-detect, robustness under attack, computational overhead and analyst-perceived usefulness.

The principal contributions are: a taxonomy and critical synthesis of AI-driven cybersecurity approaches; a reference architecture and formal threat model for intelligent defence; a hybrid detection methodology with drift and adversarial resilience; an explainability pipeline aligned with operational security practice; and an empirical evaluation protocol that emphasises realism and reproducibility. The thesis concludes with implications for practice, governance and ethics, and a research agenda for trustworthy, AI-enabled cyber defence.

The keywords listed are: artificial intelligence; machine learning; deep learning; cybersecurity; intrusion detection; threat detection; anomaly detection; adversarial machine learning; explainable AI; federated learning; automated response; MITRE ATT&CK.

Keywords

Files (1)

Cite this record

Namachila, D. (2025). Artificial Intelligence-Driven Cybersecurity: Developing Intelligent Frameworks for Advanced Threat Detection and Prevention. Life of Research. https://lifeofresearch.com/articles/199

APA 7th edition
Export to a reference manager
BibTeX
@misc{namachila2025artificial,
  author       = {Namachila, Dominic},
  title        = {{Artificial Intelligence-Driven Cybersecurity: Developing Intelligent Frameworks for Advanced Threat Detection and Prevention}},
  year         = {2025},
  month        = mar,
  date         = {2025-03-05},
  publisher    = {Life of Research},
  url          = {https://lifeofresearch.com/articles/199},
  abstract     = {Here is the abstract as it appears in the thesis:

The accelerating scale, speed and sophistication of cyber attacks have outpaced the capabilities of signature-based and rule-driven security controls. Advanced persistent threats (APTs), polymorphic malware, zero-day exploits, and increasingly AI-assisted adversaries operate across cloud, enterprise, Internet of Things (IoT) and operational-technology environments, producing volumes of telemetry that human analysts cannot triage in time. Artificial intelligence (AI), and machine learning (ML) in particular, offers a path towards security systems that learn normal behaviour, generalise to previously unseen attacks, and respond at machine speed. However, existing AI-based security research is frequently limited by unrealistic evaluation, dataset obsolescence, high false-positive rates, poor explainability, vulnerability to adversarial manipulation, and weak integration with operational workflows.

This thesis proposes, implements and evaluates an intelligent, modular framework for advanced threat detection and prevention that addresses these limitations in an integrated manner. The framework combines (i) a multi-source data ingestion and feature-engineering layer; (ii) a hybrid detection engine that fuses supervised, unsupervised and graph-based learning to identify both known and novel threats; (iii) an adversarial-robustness and drift-management layer that hardens models against evasion and poisoning and maintains performance under concept drift; (iv) an explainability and analyst-assistance layer that produces human-interpretable evidence mapped to the MITRE ATT\&CK knowledge base; and (v) an automated, policy-governed prevention and response layer that supports human-in-the-loop containment.

The framework is evaluated through a mixed-methods design comprising benchmark experiments on contemporary public datasets, cross-dataset and temporal generalisation tests, adversarial robustness assessment, ablation studies, latency and resource profiling, and expert evaluation by security practitioners. Results are reported in terms of detection performance, false-alarm rate, time-to-detect, robustness under attack, computational overhead and analyst-perceived usefulness.

The principal contributions are: a taxonomy and critical synthesis of AI-driven cybersecurity approaches; a reference architecture and formal threat model for intelligent defence; a hybrid detection methodology with drift and adversarial resilience; an explainability pipeline aligned with operational security practice; and an empirical evaluation protocol that emphasises realism and reproducibility. The thesis concludes with implications for practice, governance and ethics, and a research agenda for trustworthy, AI-enabled cyber defence.

The keywords listed are: artificial intelligence; machine learning; deep learning; cybersecurity; intrusion detection; threat detection; anomaly detection; adversarial machine learning; explainable AI; federated learning; automated response; MITRE ATT\&CK.},
  keywords     = {artificial intelligence; machine learning; deep learning; cybersecurity; intrusion detection; threat detection; anomaly detection; adversarial machine learning; explainable AI; federated learning; automated response; MITRE ATT\&CK},
  language     = {en},
  copyright    = {Creative Commons Attribution 4.0}
}

Using Zotero or Mendeley? Their browser connector saves this record from this page, with its PDF when there is one. On search and browse pages it lists every result so you can save many at once.

Guides What a complete citation needs What CC BY 4.0 allows