Evaluation of Artificial Neural Network and Multiple Nonlinear Regression Modeling for the determination of Dissolved Organic Carbon

Monitoring of organic carbon in water resources is a critical quality index in environmental management, water quality monitoring and drinking water projects. In this study, the performance and applicability of artificial neural network and multiple nonlinear regression modeling were investigated and optimized for the prediction of dissolved organic carbon. Optimization was performed using backward elimination method with the highest probable correlation coefficient and minimum number of input parameters. Model verification showed a good agreement between the predicted organic carbon and actual observations. Results showed the acceptable performance of neural network model with the mean absolute error percentage of 7.6% and correlation coefficient of 0.91. Further investigations unveiled that although the multiple regression model, with mean absolute error percentage of 8.4% and correlation coefficient of 0.89, seems to be less appealing but its fast run-time and better performance in critical conditions makes it a better choice for the prediction of organic carbon in aqueous solotions with high range of qualitative changes.