For calamity risk reduction and successful preparedness, timely and precise flood forecast is essential. Kerala is one region in southwest India that has seen several floods. Thus, the region has been selected as the research area as a consequence. In order to improve flood forecasting’s accuracy and resilience, this study looked at the use of ensemble machine learning approaches. This time, we employed a number of ensemble approaches, including voting, stacking, boosting and bagging. In this method, we employed well-known classifiers such as Support Vector Classifier, Decision Trees, KNN, XGBoost, Gradient Boost and Logistic Regression, Extra Tree Classifier. We also use hyperparameter tweaking and 10-cross-validation to identify the optimal values. During the data prepressing phase, we use data balancing techniques to address class imbalances, feature selection with a correlation matrix to identify pertinent variables, normalization to provide consistent data scaling and isolation forests to increase the accuracy of outlier detection. Additionally, PCA is utilized to extract features in order to amplify the model’s efficacy. In order to analyse the prediction %, we compute performance matrices such as ROC_AUC, F1-score, accuracy, precision and recall. With Random Forest, Extra Tree and XGBoost, the highest accuracy of 0.9778 is achieved. In flood-prone areas, the ensemble model’s enhanced predictive powers highlight its potential for real-time flood forecasting, offering vital information for emergency action and well-informed decision-making.