Published Paper Details:

Phishing has emerged as a significant cybersecurity challenge, evolving across multiple communication platforms such as SMS, email, instant messaging apps, voice calls, and malicious web links. To address this growing concern, this paper presents a unified and modular phishing detection framework capable of analyzing diverse data types including textual content, URLs, call transcripts, and mobile metadata. The system integrates natural language processing with traditional and ensemble machine learning algorithms--such as BERT, LSTM, Random Forest, and XGBoost--alongside advanced feature selection techniques like PCA and RFE to enhance detection accuracy and efficiency. Each data input type is routed through a dedicated pipeline designed to extract semantic, structural, and behavioral patterns indicative of phishing attempts. Additionally, the model incorporates image-based website analysis and speech-to-text interpretation for detecting visually deceptive pages and voice scams. Evaluated on a curated dataset of over 30,000 instances, the proposed system achieves strong detection performance while maintaining low false positive rates. The real-time processing and centralized threat visualization make it suitable for deployment in both enterprise networks and consumer-facing applications. This research contributes to scalable, cross-platform phishing prevention through intelligent, adaptable detection mechanisms.