Tools
Tools
Ponduru, Hemanth (2025) Lightweight Deep Learning Approaches for Efficient Detection of Cyber-Attack Text. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (688kB) | Preview |
Preview |
PDF (Configuration Manual)
Download (588kB) | Preview |
Abstract
The massive proliferation of connected devices across the Internet of Things, Industrial Control Systems, and social media platforms has significantly increased vulnerability to advanced cyberattacks, including malware, phishing, and botnet intrusions. Traditional rule-based systems and existing machine learning or deep learning models, despite their detection capabilities, often demand high computational resources and are unsuitable for real-time or resource-limited environments. These limitations underscore the need for models that maintain strong predictive performance while operating efficiently on constrained infrastructures. To address this challenge, this study investigates lightweight attention-based models for efficient and cyberattack text detection. The research leverages a compact LSTM-based neural architecture enhanced with a single attention mechanism that identifies the most relevant threat indicators while minimizing unnecessary computation. This selective attention enables effective operation within IoT and edge environments, where storage, processing power, and latency are critical constraints. The study implements a full pipeline encompassing data preprocessing, feature extraction, attention-driven weighting, and lightweight classification. Performance evaluation using accuracy, precision, recall, F1-score, and computational efficiency demonstrates that the lightweight LSTM + Attention model achieves superior results, reaching 94.31% accuracy and an F1-score of 94.33%, outperforming all traditional ML and standalone DL models. Overall, this study contributes of scalable, reliable, and resource-efficient cybersecurity solutions capable of adapting to the evolving complexity of modern cyber threats through the use of attention-enhanced lightweight architectures.
Actions (login required)
 |
View Item |