Publication

Abstract : Phishing attacks and deepfake media manipulation create serious and growing threats to online security, public trust, and digital integrity. This study focuses on developing a robust architecture to address both threats in a scalable, ethical, and explainable manner. The proposed framework includes two subsystems: PhishCrawl, a distributed phishing URL detection web crawler using Celery for task parallelism and Redis for shared memory, and DeepTrace, a real time deepfake detection system using a quantized Xception model. PhishCrawl employs heuristics derived from URL formats, domain characteristics, and content trends. Ethical testing is guaranteed by using mock phishing URLs. DeepTrace employs Grad-CAM for clarity and pHash for identifying perceptual similarity, implemented through a client-server model utilizing FastAPI. DeepTrace reaches a classification a ccuracy of 91.4 % ont he C eleb-DF v 2 dataset, with an average inference duration of under 200 ms for each frame. The pHash-based matching module achieves a 98 % recall in identifying reused content. PhishCrawl effectively classified phishing patterns in a distributed environment using lightweight rule-based algorithms. Together, these results demonstrate that distributed crawling with explainable deepfake detection offers a practical and reliable path to safeguarding digital ecosystems from modern cyber threats.