Comparative Analysis of Machine Learning Techniques for Predicting Nitrate Concentration in Groundwater

This study focuses on water quality assessment in the Dharmapuri district, specifically within the Karur taluk region, with a particular emphasis on total nitrogen concentrations. Water samples were collected and analyzed for total nitrogen content, and the outcomes were compared to the Water Quality Index (WQI) and World Health Organization (WHO) criteria. The optimal WQI value for water quality is 100, whereas the WHO limit for total nitrogen in drinking water is $10 ~\mathrm{mg} / \mathrm{L}$. The global challenges posed by poor drinking water quality necessitate urgent attention due to the impacts of developmental activities, pollution, and climate change on water resources. Different levels of contamination were highlighted by the analysis, which showed values ranging from $5 ~\mathrm{mg} / \mathrm{L}$ in Mampatti to $180 ~\mathrm{mg} / \mathrm{L}$ in Elavadai. Various machine learning models, including, Random Forest, Support Vector Machine (SVM), Gradient Boosting, and Neural Networks were employed for predicting nitrate concentrations. Comparative analysis revealed that while Neural Network and SVM models tended to underestimate, especially at higher concentrations, Gradient Boosting and Random Forest models produced predictions that were more in line with real nitrate levels. These patterns were supported by scatter plots comparing real and predicted values for SVM and neural network models, which showed the methods’ limits in capturing the whole range of nitrate variability. This research highlights the need for enhanced prediction models to guarantee efficient management of groundwater quality and focused approaches to reduce nitrate pollution in the area.