International Journal of Science and Technology (IJST)

Explainable Behavioral Biometric Authentication (XBBA): A Deep Learning Framework for Real-Time Phishing Detection with Interpretable AI

Phishing attacks increasingly evade traditional security measures due to their reliance on opaque AI models. We propose Explainable Behavioral Biometric Authentication (XBBA), a deep learning framework that detects phishing in real time using behavioral biometrics (mouse movements, keystrokes, scroll patterns) while providing interpretable AI explanations. XBBA employs an LSTM network with attention mechanisms to analyze user interactions, flagging anomalies like erratic cursor movements. Unlike black-box systems, it integrates SHAP and LIME to generate actionable insights (e.g., "92% phishing likelihood due to abnormal hyperlink dwell time"). Evaluated on real-world data, XBBA achieves a 95.2% F1-score, <2% false positives, and sub-100ms explanation latency—outperforming tools like Darktrace. The framework addresses critical challenges: compliance with GDPR’s "right to explanation" and building trust in SOC analysts via auditable decision trails. Key contributions include: Transparent accuracy: Fusion of biometrics and XAI. Real-time deployment: Lightweight edge-compatible design (e.g., browser extensions). Industry validation: Tested against enterprise threats. Future work extends XBBA to mobile environments and adversarial evasion scenarios. Its modular design enables integration with SIEM platforms, offering a scalable upgrade to phishing defenses.