AI-Driven Intrusion Detection System for IoT Network Security

Thesis Abstract

The rapid proliferation of Internet of Things (IoT) devices has significantly enhanced interconnected environments but concurrently heightened vulnerabilities to various security breaches, necessitating the development of robust intrusion detection mechanisms tailored specifically for IoT networks. Despite ongoing advancements, traditional intrusion detection systems (IDS) often fall short in addressing the unique challenges of IoT environments, such as resource constraints, heterogeneity of devices, and evolving threat landscapes. This study aims to design, implement, and evaluate an Artificial Intelligence (AI)-driven intrusion detection system that leverages machine learning techniques to improve real-time detection and classification of network intrusions within IoT ecosystems. The specific objectives include identifying optimal AI algorithms suited for IoT intrusion detection, developing a lightweight yet effective model capable of operating within resource-limited devices, and assessing the model’s accuracy, precision, recall, and computational efficiency. The research adopts a mixed-methods approach, combining quantitative data analysis with qualitative insights. The study’s quantitative component involves collecting network traffic data from a simulated IoT environment comprising 150 diverse IoT devices, including sensors, actuators, and gateway nodes, over an 8-week period. Data collection is facilitated through custom-developed traffic monitoring tools integrated with the IoT network, capturing both normal operations and various attack scenarios (e.g., denial of service, data injection, spoofing). The dataset, comprising approximately 1 million labeled network instances, is subsequently analyzed using supervised machine learning algorithms such as Random Forest, Support Vector Machine (SVM), and Deep Neural Networks, to determine their detection capabilities. Feature selection techniques like Recursive Feature Elimination (RFE) enhance model performance and interpretability. Qualitative insights are gathered through expert interviews with cybersecurity specialists and IoT system developers, focusing on practical deployment challenges and constraints. The primary data analysis employs regression analysis to compare classifier performance, supported by cross-validation techniques to mitigate overfitting. Model effectiveness is evaluated using metrics including accuracy, precision, recall, F1-score, and computational latency, while statistical significance of results is assessed through Analysis of Variance (ANOVA). Key findings are anticipated to demonstrate that AI-based models, particularly deep learning architectures, outperform traditional rule-based IDS in detecting complex and subtle IoT-specific threats, with an expected accuracy exceeding 90% and false-positive rates below 5%. The study also expects to identify optimal feature sets and algorithm configurations for resource-constrained devices, facilitating practical implementation in real-world scenarios. Furthermore, the research is expected to reveal critical insights into the trade-offs between detection efficacy and computational efficiency, informing best practices for deploying AI-driven IDS in IoT settings. This thesis significantly contributes to existing knowledge by providing a comprehensive framework for designing intelligent, adaptive intrusion detection systems tailored for IoT environments. It advances theoretical understanding by integrating AI and cybersecurity principles, grounded in frameworks such as the Information Processing Theory and the Theory of Anomaly Detection. The developed model offers a scalable, efficient, and accurate solution for enhancing IoT network security, addressing the gap in current detection methods that lack customization for IoT-specific constraints. The main conclusions underscore that AI-powered intrusion detection markedly enhances IoT security posture, balancing accuracy and computational resource demands. Recommendations include deploying lightweight AI models at edge devices, integrating automated threat mitigation protocols, and fostering ongoing model retraining using real-time data to adapt to emerging threats. Future research directions suggest exploring federated learning approaches for distributed detection in large-scale IoT networks and investigating the integration of blockchain technology to strengthen data integrity and security assurance in AI-driven intrusion detection systems.