<![CDATA[Rainfall prediction remains a critical challenge in meteorological science due to the non-linear and non-stationary nature of weather data. A persistent obstacle in building accurate rainfall classifiers is class imbalance, where no-rain observations significantly outnumber rain events, causing models to underperform on the minority Rain class precisely the class of greatest practical importance for flood preparedness and agricultural management. This study proposes an integrated machine learning pipeline combining feature engineering, Synthetic Minority Over-sampling Technique (SMOTE), Random Forest-based feature selection, hyperparameter tuning via Randomized Search Cross-Validation, and a soft voting ensemble of Random Forest and XGBoost for rainfall prediction. The WeatherAUS dataset containing 142,193 daily observations from 49 cities across Australia. Experimental results demonstrate that the baseline soft voting ensemble achieved the best overall performance with an Accuracy of 0.8498, ROC-AUC of 0.8800, and Weighted F1-Score of 0.8467, outperforming both standalone Random Forest and XGBoost across all metrics. Furthermore, the study finds that preprocessing quality — specifically the integration of feature engineering, SMOTE balancing, and feature selection — has a greater influence on model performance than hyperparameter optimization. This research contributes an evidence-based integrated preprocessing framework for imbalanced meteorological classification that advances the application of ensemble machine learning in rainfall prediction, with practical implications for early warning systems and data-driven weather forecasting in the field of Informatics and Computer Science.]]>
