Performance Analysis of Artificial Neural Network Models in Water Level Forecasting

Abstract

To examine the performance of Artificial Neural Networks (ANNs) in predicting the water level of a watershed three days ahead of time, this study evaluated eighteen ANN models with different combinations of training algorithms and activation functions. The two best models corresponding to the two major climate seasons in the Philippines were: for the rainy season; it was the Resilient Propagation with Leaky ReLU which produced MAPE and RMSE of 6.731 and 0.0098, respectively; for the dry season, it was the Scale Conjugate Gradient with Leaky ReLU which produced MAPE and RMSE of 7.871 and 0.01045, respectively.