<![CDATA[Purpose: This study aims to identify the right combination of network architecture, learning rate, and epoch in making predictions at each rainfall post in Maros Regency. In addition, this study also predicts the monthly rainfall profile in 2021-2025 in Maros Regency.Methods: The method in this study is the backpropagation neural network algorithm to learn and predict the data. BPNN is one of the most commonly used non-linear methods in making predictions recently. The data used in this study is monthly rainfall data from 2000-2020 as training and testing data at four rainfall stations including BPP Batubassi, Staklim Maros, Stamet Hasanuddin, and BPP Tanralili.Result: The results showed that the combination of network architecture, learning rate, and epoch obtained at each rainfall post was different. The highest level of prediction accuracy was obtained on 5 layers rather than 3 or 4 layers of network architecture with prediction accuracy at each rainfall post respectively 76.91%, 72.47%, 75.24%, and 76.53%. The predictions of rainfall from 2021-2025 are following the monsoon rain pattern with the highest rainfall in January 2025 of 964.1 mm, while the largest annual rainfall is obtained in 2023 with a total of 3359.6 mm.Novelty: In this study, various combinations of network architecture parameters consisting of learning rate, epoch, and architecture at each rainfall post obtained different results. Particularly in the Maros Regency, the combination that is most suitable for use in predicting monthly rainfall at the Batubassi BPP post is learning rate 0.7, epoch 50000, and network architecture 11-6-10-7-5, at Staklim Maros post is learning rate 0.5, epoch 50000, and network architecture 11-5-9-10-5, at Stamet Hasanuddin post is learning rate 0.8, epoch 20000, and network architecture 11-5-8-6-5, and at BPP Tanralili post is learning rate 0.5, epoch 10000, and 11-5-9-9-5 network architecture.]]>
Copyrights © 2023
