Estimating Contamination Risk Using Artificial Intelligence Models a Case of the Patiño Aquifer, Paraguay

Abstract

Studying the risk of contamination is essential for the protection of aquifers. In Paraguay, one of the major drinking water supplies is the Patiño Aquifer. A previous study, using a deterministic model, identified that 42% of the aquifer have a high risk of contamination. This work uses artificial intelligence (AI) models, with regression and classification approaches, to estimate the contamination risk of the urban zone of the Patiño aquifer by Total Nitrogen (TN). The Supervised Committee Machine with Artificial Intelligence (SCMAI) model is applied as a regression model, which combines the Artificial Neural Network (ANN), Mamdani Fuzzy Logic (MFL), Sugeno Fuzzy Logic (SFL) and Neuro Fuzzy (NF) models. Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and the correlation of estimated risk with TN concentration are used for the evaluation. The Decision Trees (DT), Bayesian Network (BN) and K-Nearest Neighbor (KNN) models are used for the low and not low risk classification approach. These models were evaluated based on the precision, recall and accuracy indicators. The SCMAI model improved the performance and correlation of the ANN, MFL and SFL models, with RMSE, MAE and correlation values of 2.33, 1.38 and 0.86 respectively. The J48 and PART algorithms, applied to the DT model, and the KNN model obtained an accuracy of 99%, and the BN model obtained an accuracy of 98%. Precision and recall values showed that the DT algorithms failed less, by 10%, and that it is able to identify 81% of the cases of the not low risk levels. It was observed that both approaches give a competent picture of the state of the Patiño aquifer in relation to the risk of contamination.