Abstract
Sorting and analyzing different types of rainfall according to their intensity, duration, distribution, and associated meteorological circumstances is the process of rainfall prediction. Understanding rainfall patterns and predictions is crucial for various applications, such as climate studies, weather forecasting, agriculture, and water resource management. Making educated decisions about things like agricultural planning, effective use of water resources, precise weather forecasting, and a greater comprehension of climate-related phenomena is made more accessible when many components of rainfall are analyzed. The capacity to confront and overcome this obstacle is where machine learning and metaheuristic algorithms shine. This study introduces the Adaptive Dynamic Particle Swarm Optimization enhanced with the Guided Whale Optimization Algorithm (AD-PSO-Guided WOA) for rainfall prediction. The AD-PSO-Guided WOA overcomes limitations of conventional optimization algorithms, such as premature convergence by balancing global search (exploration) and local refinement (exploitation). This effectively balances exploration and exploitation, and addresses the early convergence problem of the original algorithms. To choose the most crucial characteristics of the dataset, the feature selection method employs the binary format of AD-PSO-Guided WOA. Next, the desired features are trained on five different models: Decision Trees (DT), Random Forest (RF), Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), and K-Nearest Neighbor (KNN). Out of all the models, LSTM produced the best results. The AD-PSO-Guided WOA algorithm was used to adjust the hyperparameters for the LSTM model. With coefficient of determination (R2) of 0.9636, the results demonstrate the superior efficacy and performance of the suggested methodology (AD-PSO-Guided WOA-LSTM) compared to other alternative optimization techniques.