Abstract
Flood is a natural phenomenon that occurs without warning. The occurrence of flood can cause many damages in terms of property, money, and even death. In order to prevent more loss, an early warning system of the increment of river water level needs to be forecasted in a more accurate and effective way to predict the best result. In this study, Artificial Neural Network is implemented in order to forecast the river water level in Pelarit River, Perlis. The objectives of this study are to determine the efficiency of ANN in forecasting the river water level and to forecast the water level of Pelarit River for one week in advance. This study uses Multilayer Feed Forward Artificial Neural Network with two input neurons, seven hidden neurons, and one output neuron in order to forecast the water level. Three different types of algorithms, namely quick propagation, conjugate gradient descent, and Levenberg–Marquardt algorithms, are compared to get the most significant result in forecasting one week of river water level. Among all the algorithms used, conjugate gradient descent is proved to be the most reliable value to forecast the Pelarit River in terms of the lowest value of Root-Mean-Square Error and the highest value of correlation.