Identification of significant rainfall stations in Kelantan River using Z-score for Multi-Layer Perceptron (MLP) model development

Abstract

As part of ongoing research for flood modeling, this paper proposes Z-score in identifying the significant rainfall stations in Kelantan River for Multi-Layer Perceptron (MLP) model development. In this work, time series data of river water level was collected with hourly time interval in year 2015. Seven rainfall stations in Kelantan River were used and applied to Z-Score. They are Gunung Gagau 1, Kuala Koh, Tualang, Kuala Krai, Kusial, Dataran Air Mulih and Jeti Kastam, and coded as S1 to S7, respectively. The result showed that rainfall station S2, S3 and S5 were identified and the river water level of these station are fed to Multi-Layer Perceptron (MLP) for model development to predict the river water level at station S5. The input and output data is splitted to training, validation and testing with the ratio of 70%:15%:15%. The neurons in hidden layer were varying from 1 to 10 and Mean Square Error (MSE) for each neuron is computed together with regression for training network. It showed that the regression for training network is 0.9999 which is closed to 1 accompanied by the smallest MSE is 0.00027 at 3 hidden neurons in hidden layer. The finding in this study revealed the capability of Z-Score in identifying significant rainfall stations and MLP developed model successful in predicting river water level at station S5.