mdsajjadullah/bangla-phishing-detection-2026

Bangla Phishing Detection Dataset (SMS, Email, URLs) 2026

Synthetic dataset (~2000 rows) of phishing and legitimate messages in Bangla (Bengali) + some English, simulating common Bangladesh scams (bKash, Nagad, Daraz, Eid offers, job fraud, account lock alerts, etc.).

Research Motivation

Phishing/smishing attacks are rising in Bangladesh and South Asia, often in Bangla using local services. Most phishing datasets are English-only and miss these patterns.
This dataset fills the gap for Bangla NLP models, local detectors, and cybersecurity education in low-resource languages.

Columns

• text: full message (SMS body, email subject+body, or URL)
• type: sms / email / url
• label: 1 = phishing, 0 = legitimate
• language: bn / en / mixed
• source: simulation info
• english_translation: English version (partial in some rows)
• has_urgency: 1 if urgency keywords detected
• has_shortened_url: 1 if shortened URL detected

Stats

• Rows: ~2000
• Phishing: ~65%
• Types: ~50% SMS, 30% email, 20% URL

Creation

• 100% synthetic (Faker + custom Bangla templates)
• No real user data used
• Baseline: TF-IDF + Logistic Regression (~90%+ accuracy on synthetic data — see notebook)

Intended Use

• Fine-tune BanglaBERT / similar for phishing detection
• Educational cybersecurity / NLP projects
• Research on low-resource fraud detection

Important Note

Simulated data only — for research/education/ethical simulations.
Not for production without real-world validation.

Feedback welcome! v2 planned with full translations.

Created by Md.Sajjad Ullah (Dhaka, Bangladesh)