Natural disaster, particularly floods being the most devastating kind pose significant threat to human society. With the increasing frequency of Natural disaster such as floods and the growing complexity of environmental factors, flood prediction have become a very critical challenge. In this study we’ve performed a exploratory analysis on large scale flood probabilities datasets that has 21 significant attributes such as Monsoon Intensity, Deforestation, River Management, and Urbanization. We compare the performance of traditional machine learning algorithms such as Multiple Linear Regression, K-Nearest Neighbors, Lasso Regression, Ridge Regression, and Multi-layer Perceptron (MLP) and improve the accuracy of prediction with a hybrid stacking model. The stacking model proposed in the work is a blend of XGBoost and MLP with the meta-learner being an XGBoost Regressor. The results reveal that the hybrid model is superior to standalone models achieving R \(^{2}\) = 0.8612 and MSE = 0.00036, the evaluation was carried out on a 80/20 train-test split, validated through grid-searched hyperparameters and standardized preprocessing. The Stacked model proved to be a promising and applicable technology in actual flood risk management and disaster prevention and response.