Phishing continues to be a significant and evolving threat to cybersecurity. The current methods for detection are not up to par when managing new and dynamic attacks such as typo squatting and evil redirects. This research paper proposes an intelligent phishing detection framework that combines Random Forest-based Weighted Bootstrap Sampling (WBS), iterative LASSO feature selection, and real-time processing to achieve high performance and adaptability. Many of the existing systems are deficient in real-time capabilities, transparency, and resilient mechanisms against newly emerging threats. The proposed model can therefore fill this gap by working on URL analysis in real-time and providing explanations by using SHAP and LIME. The system relies on Random Forest as the base classifier, with LASSO assisting in iterative feature selection and WBS in efficient back-testing. Through this methodology, an effective detection system is created that detects attempts of phishing along with providing interpretable insights and real-time alerts on threats. Upon evaluation with 50,000 URLs (half phish, half legitimate), the model hit 98.2% accuracy, 1.4% false positive rate, and 97.8% phishing detection. Real-time analysis had elevated detection capabilities for typo squatting, and explainers increased trust in the system. The alerts also enabled a 45% faster reaction from the users in threatening situations. Marrying AI with interpretable and proactive techniques results in a transparent and well-refined phishing detection system.