Rainfall Forecasting Using Hybrid Learning Method

Rainfall is one of the most important meteorological variables. Rainfall has a multiplicity of effects on agriculture, ecosystems and urbanization. The amount of rainfall cannot currently be determined due to climate change. For this reason, in order to mitigate the effects that arise from heavy rainfall, it is crucial to forecast the amount of rainfall. Currently, statistical techniques and machine learning are two of the methods has been used to forecast rainfall. These methods have been successfully to forecast rainfall. However, it still has low accuracy and need more time to forecast because the rainfall influence parameters are time-series and non-linear. This study will employ a hybrid learning technique to resolve the issue. The method used is a combination of two algorithms, namely long short-term memory (LSTM) and particle swarm optimization (PSO), and the prediction system will be evaluated using RMSE and MAE. Time-series and non-linear rainfall data will be overcome by the LSTM algorithm, and uncertain hyperparameter settings in the LSTM method will be overcome by the PSO algorithm. The ideal LSTM algorithm employing PSO optimization yields the following hyperparameter value results: 143 neurons, learning rate of 0.0000840, and 122 epochs in 50 iterations. With the lowest RMSE and MAE values of 18.557095 and 8.972361, respectively, the prediction model's optimal hyperparameter values can forecast the next five days based on the data used.