<p>Flood prediction is a crucial aspect of disaster management, requiring accurate forecasting to minimize damages and enhance preparedness. Traditional models often fail to capture both spatial and temporal dependencies in environmental parameters, leading to suboptimal predictions. To address this, we propose a hybrid CNN-GRU-Dense model that leverages Convolutional Neural Networks (CNN) for spatial feature extraction, Gated Recurrent Units (GRU) for sequential learning, and Dense layers for refined feature representation named CGD-FloodNet. The model is trained and evaluated on a real-time environmental dataset incorporating rainfall, temperature, water levels, and humidity. Experimental results demonstrate the superior performance of the proposed model, achieving the lowest Mean Squared Error (MSE) of 0.0021, Root Mean Squared Error (RMSE) of 0.0458, and Mean Absolute Error (MAE) of 0.0304, along with the highest coefficient of determination (R²) of 96.52% and Explained Variance Score (EVS) of 96.71%. A thorough statistical test analysis, including paired t-tests, Wilcoxon Signed-Rank tests, and ANOVA, confirms the significant improvement of our model over baseline architectures such as CNN-only, GRU-only, LSTM, and ANN. Comparative evaluation with state-of-the-art methods highlights its robustness in flood forecasting applications. The proposed model provides an efficient and reliable solution for real-time flood risk assessment, offering valuable insights for disaster preparedness, early warning systems, and climate resilience strategies.</p>