Metaheuristic Optimization Stacking Application for Rainfall Classification: A Comparative Study

Accurate rainfall classification plays a vital role in effective meteorological forecasting, agricultural planning, and also to avoid natural disasters. However, standard classification models often exhibit unstable performance. This study evaluates the effectiveness of an Ensemble Stacking framework enhanced with Optimized Swarm Based and Metaheuristic method such as Artificial Bee Colony (ABC) and Cuckoo Search (CS) optimization algorithms to improve prediction reliability. The proposed approach was tested using a detailed rainfall dataset by combining basic classification models, such as Decision Tree, SVM, Naive Bayes, and kNN. The results show that Stacking Ensemble generally outperform individual basic models in Accuracy and F1 Score (reaching a median > 0.80), while unoptimized Stacking method show low variances but provides a less better result in terms of Accuracy and F1 Score. In contrast, the Stacking model optimized with ABC emerged as a better method, demonstrating the highest stability and significantly reducing the performance distribution range compared to the non-optimization and Cuckoo Search scenarios. These findings conclude that the application of Artificial Bee Colony optimization to Stacking ensembles effectively minimizes prediction variance, making it the most reliable strategy for consistent rainfall forecasting using classification modelling technique.