Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{204237,
        author = {Ram Gopal},
        title = {A Compute-Efficient Hybrid Framework for Phishing Email Detection Using Frozen DistilBERT and Interpretable Features},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {1},
        pages = {2558-2563},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=204237},
        abstract = {Phishing emails continue to be a significant cybersecurity risk, advanced transformer-based detection systems demand considerable computational power, hindering implementation on email gateways with limited resources to combat these attacks. This research presents a compute-efficient hybrid structure that integrates frozen DistilBERT embeddings combined with interpretable custom features 21-dimensional and TF-IDF. representations, handled via a streamlined classification head. The icy transformer backbone reduces gradient flow through transformer layers, minimizing training complexity while maintaining transferable semantic representations acquired during pretraining. Experiments are performed on 39,747 training emails sourced from the Enron and SpamAssassin dataset employing stratified 80/20 divisions. F1 = 0.990 in in-domain validation, maintaining the performance of fully fine-tuned DistilBERT while decreasing training duration decreased by 60% and maximum GPU memory usage reduced by 38%, optimized variants which attain merely 33.5% precision and 0.343 balanced accuracy during domain shift. These results suggest that maintaining transformer representations unchanged while incorporating domain- specific handcrafted heuristics creates a more resilient and practical phishing detector for deployment than complete fine-tuning.},
        keywords = {Email security, Frozen transformers, Hybrid neural architecture, Phishing detection.},
        month = {June},
        }