Smart Phishing Detection: Integrating Neural Networks and Ensemble Learning for Enhanced Cybersecurity

Phishing attacks have emerged as one of the most unavoidable and damaging cyber threats these days, with cybercriminals unfolding new tricks to deceive and defraud users, targeting individuals and businesses alike. Hence, this paper proposes a new NN ensemble-based high accuracy Phishing website detection system combining the features of both supervised and unsupervised schemes which also provides Real-time detection, multigoal support and uses a Phishing history database. The system employs Isolation Forest for its anomaly detection, followed by Ensemble learning methods such as Random Forest, LightGBM, and Gradient boosting. Adding isolation forest anomaly score as one of the features in our ensemble classifier it works better for detection. The new model also preferred the real-time detection in phishing and supports multi-language over several languages and nations, thus completed its worldwide utility while maintaining a history of phishing by preserving a database of previous attacks for results. The experimental results demonstrate excellent performance in fighting upcoming attacks, on a variety of languages, high accuracy, and the adjustment to quasi real-time applications are the main advancements that can be taken from the present work.