The cybersecurity world faces an ongoing challenge from phishing attacks, which uses fake emails and websites to induce an attack and steal sensitive data. The increase in growth of adaptive attack techniques, together with feature duplicacy and information imbalance, creates intense challenges for traditional detection methods. This paper presents a Deep Learning Enhanced Boosting-Based Ensemble Model for Phishing Detection. This model integrates advanced boosting strategies, like XGBoost, CatBoost, and LightGBM, to optimize both feature selection and classification prediction. This hybrid feature selection system functions by combining the feature importance ranking from the most precise models. Ensemble methodology with multilayer stacking overcomes single model weaknesses through the method of boosting model integration. The Convolution Neural Network- Long Short-Term Memory (CNN-LSTM) model demonstrates superior detection capabilities through its ability to extract automatic complex patterns that identify sequential relationships between phishing data. This integrated system achieves both better performance and generalization, which delivers an effective approach for phishing detection.