Prediction of arsenic contamination in Rayong groundwater basin using machine learning based approaches

Abstract

The present study used three algorithms consisting of Support Vector Machine (SVM), Random Forest (RF), and Artificial Neuron Network (ANN) to locate risk area of arsenic (As)contamination in Rayong coastal aquifers, Thailand. There were three parts in this study consisting of 1) selecting the proper parameters 2) selecting the appropriate model, and 3) constructing the risk map of As. To perform models efficiently, the parameters used to generate the models have to be selected based on the correlation of each hydrochemical parameter with As concentration, which could explain the mechanisms of As release in groundwater. Due to major parameters in the dataset were monotonic and not presented by the normal distribution, thus, Spearman’s correlation was conducted to screen the suitable parameters. The results showed that parameters correlated with As mostly supported by the mechanism of As release in groundwater, which is dominantly controlled by the reducing condition. Spearman’s correlation technique would help to select the crucial parameters in the further modeling process. To select an appropriate model to generate the risk map, the model’s performance has to be measured by the prediction performance and uncertainty of each model. The prediction performance indicated that the RF algorithm has the highest performance as compared to those in SVM and ANN. In addition, the uncertainty of each model confirmed that the RF algorithm has the lowest uncertainty. Moreover, to confirm the performance of the models, the actual As concentration in field data were used to validate the prediction result of each model. The result, also confirms that the RF model was the best performance model compared with the other two models. Therefore, the RF was the appropriate algorithm that can generate the probability map to locate the areas of As contamination in groundwater. The result of the risk map obtained from the RF model indicated that the deep aquifer (granite aquifer, Gr), in the northern part of the Rayong basin has a higher risk for people who have used groundwater to expose to As. In contrast, the shallow aquifer revealed that the southern part of the Rayong basin has a higher risk for people who use groundwater, which is also supported by the location of the landfill and industrial estates in the Mueang District. The outcome of this study can be useful for the government and other organizations for groundwater resource management and environmental protection. Furthermore, the novelty of this research can be used to further study other groundwater aquifers contaminated with As in the world.