Predicting Nitrate Levels in the Saïss Water Table: A Comparative Study of Machine Learning Methods

Abstract

The main goal of this study is to predict nitrate (NO3-) levels in the Saiss basin water table as a function of various physicochemical parameters. To accomplish this, three machine learning approaches were utilized: multiple linear regression (MLR), super vector regression (SVR), and artificial neural networks (ANN). The independent variables were composed of six water quality parameters, including Ca2+, Na2+, EC, Cl-, HCO3-, and SO42-. The study utilized a dataset of 389 water samples collected between 1991 and 2017. The artificial neural network (ANN) was trained using the Levenberg-Marquardt (LM) algorithm, which was selected from various optimization algorithms. Additionally, during the training of the SVR model, it was observed that the RBF kernel outperformed the other kernels (linear, polynomial, and sigmoid kernel). The results were analyzed by the coefficient of determination (R2) and the mean square error (MSE). The results of the MLR method revealed R2 (0.523) and MSE (757.34). The ANN model with architecture [6-20-1] performed better than RLM with R2 = 0.836, MSE= 0.023 The SVR model result confirms what has been proved by ANN concerning the performance, with R2=0.902 and MSE= 4,364.