Deep learning-based phishing classification framework for accurate detection using optimized URL intelligence

The proliferation of internet services has exposed customers to phishing attempts that steal sensitive information via false URLs. Intelligent categorization systems are required to tackle dynamic phishing techniques, which defy rule- and signature-based detection. The existing phishing detection systems leverage handmade characteristics or fixed blacklists. That suggests they generalize poorly on zero-day and camouflaged phishing URLs. High false-positive rates and inadequate scalability hamper their performance. It proposes an Adaptive Deep URL Intelligence Network (ADUIN). Deep learning model with optimized URL lexical, host-based, and structural properties. We optimize features using a hybrid relevance-ranking method and train a multi-layer deep neural architecture to understand complicated non-linear phishing patterns. URL intelligence dynamically updates the architecture to resist attack behavior changes. According to experiments on the benchmark phishing dataset, ADUIN is more accurate, exact, and remembers than machine learning classifiers. Zero-day phishing URLs are detected with minimal false alarm rates by the algorithm. The suggested system improves phishing URL classification accuracy, versatility, and intelligence. Real-time online and enterprise security solutions benefit. Under high load, the recommended ADUIN model has 95% classification accuracy, 93% precision, 92% zero-day detection rate, 3.5% false positives, optimal accuracy with 50 features, and 210 ms delay.