Predicting the floe characteristics of River Niger using Artificial Intelligent Models

Abstract

Artificial intelligence (AI), as a branch of computer science, is capable of analysing long-series and large-scale hydrological data. In recent years, AI technology has been applied to the hydrological forecasting modelling. It is essential to determine the hydrological system of River Niger, which is the major water sources of the annual flood in Lokoja, Kogi State, Nigeria. This paper investigates and compares the forecasting capability of three algorithms namely Artificial Neural Network (ANN), Support Vector Machine Regression (SVM Reg.) and Random Forest (RF) to determine the optimal model for forecasting downstream river flow. Daily discharges data from 2001 to 2019 were obtained from National Inland Waterways Authority at Lokoja, Kogi State, Nigeria and applied in the forecasting analysis. Discharge data were divided into 65:35 percent for training and testing respectively. The results of evaluation criteria based on Root Mean Square Error (RMSE), Nash Sutcliffe Efficiency Coefficient (NSEC), Coefficient of correlation (CC) and Accuracy (ACC) showed that all the models applied gave perfect results except the value obtained for uncertainty analysis in ANN model which was 1.4445 and 0.6219, was slightly high when compare with the values of RF 0.1634 and 0.0134 and SVM Regression models 0.1634 and 0.1210 in testing and training phases respectively. This is caused by the failure of ANN model to carry out pre-processing of discharge data, to remove all the error present in the data unlike the SVM Regression and RF models. Therefore, the RF and SVM Regression algorithms are considerably more adaptive in optimizing the forecasting problem for the river flow prediction.