An Explainable Ensemble Framework for Real-Time Phishing URL Detection

The above-mentioned survey has provided a detailed anal- ysis of the latest development of Phishing URL Detection, highlighting the shift from conventional blacklisting tech- niques to machine learning and deep learning approaches. From the analysis of recent literature, it can be concluded that approaches like XF-PhishBERT and stacking of machine learning architectures have revolutionized the performance of detection techniques. The use of XAI (Explainable AI) tools has resolved the problem of the ”black box” issue in neural network models. Nevertheless, certain problems still persist in terms of compu- tational costs and the susceptibility of the model to zero-day attacks. In future studies, the development of light-weighted temporal models must be considered. The current study presents an in-depth exploration of the current scenario of Phishing URL detection, tracing back to the revolutionary change from the static blacklisting approach and the heuristic filter technique to machine learning and deep learning models. The study of ten advanced technolo- gies in 2024-2025 demonstrates significant advancements; for instance, XF-PhishBERT achieves an accuracy rate of 99.9 The implementation of Explainable AI paradigms such as SHAP for global interpretability, LIME for local fidelity, attention heat maps, and linguistic explanations derived from LLMs has effectively addressed the ”black-box” problem in deep neural models. Benchmarking performance (see Table I) indicates that accuracy is regularly above 97 To address the remaining challenges, EEF-Phish has been developed to leverage the ensemble between LightGBM and Random Forest with XAI models in an online learning frame- work. Our current evaluation results in 97.8 Key areas that should be investigated in future research include: (1) federated zero-day learning in siloed companies; (2) adversarial hardening techniques against obfuscations; (3) Neural Architecture Search for edge devices with less than 1 MB of memory space; (4) monitoring of concept drifts with sliding-window approaches; and (5) human-centered interfaces for providing natural language threat intelligence to users. By implementing the aforementioned needs, we would be able to create unbeatable lightweight

[1]     M. S. Uddin, M. A. Kabir, and S. Islam, ”XGBoost-SHAP Based Phishing URL Detection with Explainable AI,” 2025 International Conference on Cybersecurity and Computer Science (ICCCS), IEEE, 2025. [2]     J. Chen, L. Wang, and T. Zhang, ”XF-PhishBERT: Few-Shot Learning Framework for Phishing Detection using ModernBERT,” IEEE Transac- tions on Information Forensics and Security, vol. 20, no. 3, pp. 112-125, 2025. [3]     A. Sharma, R. Gupta, and V. Singh, ”Multi-Layer Stacked Ensemble Model for Robust Phishing URL Classification,” 2025 7th International Conference on Computing, Communication and Automation (ICCCA), IEEE, 2025. [4]     P. Rao, S. K. Nayak, and G. Panda, ”SLA-FS: Hybrid Feature Selection Framework using SHAP and LIME for Phishing Detection,” 2024 IEEE 9th International Conference on Communication and Electronics Systems (ICCES), IEEE, 2024. [5]     H. Guo, Y. Liu, and X. Chen, ”EXPLICATE: Enhancing Phishing Detection Trust through LLM-Generated Explanations,” Proceedings of the 2025 ACM SIGSAC Conference on Computer and Communications Security (CCS), ACM, 2025. [6]     O. K. Sahingoz, E. B. Tuncel, and A. O. Batur, ”Attention-Augmented Bi-LSTM for Phishing URL Detection,” Computers & Security, vol. 142, 103856, Elsevier, 2024. [7]     R. Kumar, S. Gupta, and A. Sharma, ”LightGBM with LIME Explana- tions for Edge Phishing Detection,” 2024 International Conference on Intelligent Computing and Control Systems (ICICCS), IEEE, 2024. [8]     S. Alghamdi, M. Alqahtani, and A. Alharbi, ”CNN-LSTM Hybrid Architecture for Malicious URL Classification,” Springer Journal of SN Computer Science, vol. 6, no. 2, 2025. 9.       P. Verma and V. K. Shrivastava, ”Random Forest with Counterfactual Explanations for Phishing Detection,” 2024 5th International Conference on Advances in Computing, Communication and Control (ICACCI), IEEE, 2024. 10.    L. Zhang, J. Li, and H. Wang, ”Graph Neural Networks for Phishing URL Relationship Analysis,” IEEE Transactions on Network and Service Management, vol. 22, no. 1, pp. 45-58, 2025.