ABC-Optimized Deep Learning Approach to Rainfall Forecasting in Serang City

In recent years, there has been a notable increase in hydrogeological disasters, such as floods and landslides, in various regions of Indonesia, primarily attributed to the occurrence of intense rainfall events. Hence, the prediction of precipitation holds significant importance in mitigating susceptibility and the potential for inundation. The issue of rainfall forecasting is widely acknowledged as being highly challenging, and deep learning models are commonly employed to make predictions using intricate timeseries rainfall data. Nevertheless, the duration of training sessions is significantly expanding as a result of the extensive time series data and the fluctuating nature of deep learning hyper-parameter models. This paper showcases the effectiveness of the Artificial Bee Colony (ABC) method in improving the hyper-parameters of well-known deep learning models, such as the standard RNN, LSTM, BiLSTM, and GRU. The precipitation data included in this research is sourced from the Serang City Meteorology, Climatology, and Geophysical Agency (BMKG), spanning the time frame of January 2010 to December 2022. The findings indicate that the integration of ABC-optimized GRU yielded the highest level of performance, as seen by the MAE value of8.1846 and RMSE value of 13.5129. This performance surpassed that of the normal RNN, LSTM, and BiLSTM models.