Published Paper Details:

Phishing attacks persist as a critical and financially destructive threat in the digital domain, exploiting user weaknesses to acquire sensitive data such as bank credentials and login details. 1 The annual worldwide financial impact of these attacks has been estimated to be as high as US$5 billion, demonstrating the urgent need for robust detection mechanisms. 1 Traditional detection methodologies, particularly blacklisting, are inherently incapable of addressing zero-hour threats due to the attackers' frequent use of fast-flux networks and algorithmic URL generation. 1 This manuscript introduces the TrustLayer system, a novel hybrid architecture that seamlessly integrates a comprehensive, multi-tiered heuristic feature model with optimized ensemble Machine Learning (ML) classification. The synthesized heuristic model incorporates 16 distinct features, focusing on sophisticated structural evasion tactics (e.g., Punycode homograph attacks, excessive dot counts, and URL shortening services). 1 By leveraging empirically proven ensemble methods, notably the Random Forest algorithm, which achieved a detection accuracy of 97.14% and a low False Negative Rate (FNR) of 3.14% in prior studies 1 , TrustLayer is engineered to provide superior detection fidelity and resilience. The proposed system employs a staged classification pipeline optimized for low latency, ensuring real-time capability essential for mitigating modern cybersecurity threats.