TB-DBN: A hybrid deep learning architecture with IBBA Optimization for enhanced phishing URL Detection

Kavva, Sruthi Reddy (2025) TB-DBN: A hybrid deep learning architecture with IBBA Optimization for enhanced phishing URL Detection. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (846kB) | Preview
Preview	PDF (Configuration Manual) Download (669kB) | Preview

Abstract

From 2022 to 2024, phishing increased by 61%, necessitating the use of increasingly sophisticated detection techniques beyond basic blacklists. This study offers a novel hybrid structure that combines Transformer-based contextual representation and DBN hierarchical learning, which is further improved by the Intelligence Binary Bat Algorithm (IBBA). The lack of adaptive preprocessing, the unexplored hybrid transformer-DBN architectures, and the absence of metaheuristic optimisation of deep hybrid systems are the three main shortcomings that the study addresses.

The TB-DBN system enables adaptive preprocessing, which adjusts the feature extraction based on the characteristics of the URLs. 169 features that fall into eight categories are extracted. Through the use of fusion, the architectural design integrates the DeBERTa-v3-base transformer and the multi-layer DBNs, and the IBBA regularly optimises the hyperparameters.

A 96.38% F1-score (±0.0023) is obtained from the test on 11,430 URLs, significantly outperforming DBNs (81.4%) and unified transformers (87.5%). The contribution made by this work goes beyond performance indicators; it provides a unique approach to combining languages and structure to create patterns that prevent millions of dollars in losses every year. Using the same methodology across all cybersecurity domains is made easier by an optimised design through IBBA and an adaptive preprocessing framework. One significant advancement in combating constantly evolving threats is the 8.5% relative improvement over single-model approaches. An important step forward for upcoming projects and broad industry adoption is open-source software. This study shows that the future of reliable phishing detection lies not in larger models but in intelligent architectural integration.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Raj, Kislay UNSPECIFIED
Subjects:	Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics > Artificial intelligence Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Artificial intelligence P Language and Literature > P Philology. Linguistics > Computational linguistics. Natural language processing
Divisions:	School of Computing > Master of Science in Artificial Intelligence
Depositing User:	Ciara O'Brien
Date Deposited:	28 May 2026 14:31
Last Modified:	28 May 2026 14:31
URI:	https://norma.ncirl.ie/id/eprint/9325

Actions (login required)

View Item