Penerapan metode radial basis function neural network dalam memprediksi curah hujan

Abstract

North Tapanuli Regency is highly vulnerable to natural disasters such as floods and landslides due to high rainfall intensity and complex geographical conditions. To anticipate these risks, this study aims to predict rainfall using the Radial Basis Function Neural Network (RBFNN) method. The number of nodes in the input layerwas determined based on significant lags identified in the Partial Autocorrelation Function (PACF) plot. The hidden layerwas constructed using the K-Means clustering method to define cluster centers and standard deviations. Model performance was evaluated using the Root Mean Square Error (RMSE) metric. The results indicate that the optimal network structure is found in the 4-7-1 architecture, yielding an RMSE of 11.006. This model was then used to predict rainfall for the period from September 1 to December 31, 2024. The prediction results show that rainfall intensity tends to remain stable, ranging from 5 to 9 mm, with the highest value of 9.35 mm on September 1 and the lowest value of 5.40 mm on September 22. Based on these findings, the potential for natural disasters due to high rainfall during the forecast period is relatively low. The RBFNN model has proven effective for handling nonlinear meteorological data patterns, making it a valuable tool for a reference for further academic research in rainfall prediction and data clustering.