Prediction of nitrate concentrations using multiple linear regression and radial basis function neural network in the Cheliff River basin, Algeria

Abstract

In this paper, multiple linear regression (MLR) and radial basis function neural network (RBF-NN) are applied to predict nitrate (NO3 -) concentration with and without reservoir volume (WV) as predictor using monthly data for ten years in three water reservoirs located in the upper Cheliff basin (NW of Algeria). The datasets were divided into training (80%) and testing (20%) sets and two different scenarios were compared. The results revealed that RBF-NN was more efficient (MAE = 0.192 and SI = 0.061) compared with the MLR model to predict NO3 - in all reservoirs. RBF-NN provided the best accuracy in the testing period with a high R2 of 0.957 in reservoir II, and low MSE and PBias of 0.048 mg/l and 2.98% in the training period in reservoir III, respectively. Overall, the best results were generated by M(iii) in scenario B.