Evaluation of Machine Learning Models in the Prediction of Water Quality Index for Selected Water Sources in Uyo, Akwa Ibom State, Nigeria

Traditional techniques of assessing water quality using water quality index calculations are typically costly and time-consuming. With the advent of Artificial Intelligence (AI) and Machine Learning (ML) Model technologies, water quality index can be easily predicted. In this study, six machine learning models, namely the Extra Trees Regressor, Gradient Boosting Regressor, MLP, K Neighbors Regressor, LassoLarsCV and Ridge Regression were evaluated for their accuracy in prediction of groundwater quality index (GWQI) for selected water sources in Ikono, Oku, Etoi, and Offot Clans in Uyo Local Government Area towards improving the water resources management in Akwa Ibom State, Nigeria. Groundwater samples were collected and analyzed using standard methods for thirteen (13) physicochemical parameters which included: pH, temperature, electrical conductivity (EC), total dissolved solids (TDS), dissolved oxygen (DO), biological oxygen demand (BOD), alkalinity, acidity, total hardness, chlorides (Cl ⁻ ), sulphate (SO ₄ ²⁻ ), phosphate (PO ₄ ³⁻ ) and nitrate (NO ₃ ⁻ ). GWQI was calculated manually using the weighted arithmetic water quality index (WAWQI) method. Results obtained ranged from 27.21 (BH 36) to 86.81 (BH 19). The water samples were classified into three categories. Most of the samples, about 48.3%, fell into the poor water quality class, about 35% into the good water class, and 16.7% into the very poor water quality class. The experimental data was used in Machine Learning models training, testing, as well as predicting GWQI. The predicted values of GWQI were compared to the actual values from manual calculations. Evaluation of the machine learning models’ performance revealed that LassoLarsCV model was the best in predictive capability (R ² = 1.0000, RMSE = 0.00, MAE = 0.00), outperforming the Ridge Regression model (R ² = 0.9999, RMSE = 0.0862, MAE = 0.0437), the Extra Trees Regressor model (R ² = 0.9848, RMSE = 2.51, MAE = 0.65), Gradient Boosting Regressor,(R ² = 0.9920, RMSE = 1.82, MAE = 0.89), MLP (R ² = 0.9673, RMSE = 3.67, MAE = 2.19) and the K Neighbors Regressor, (R ² = 0.9516, RMSE = 0.0862, MAE = 0.0437), making it a reliable tool for accurate water quality index prediction.