Predictability Indicators of River Water Quality with AI-ML Modelling

Abstract :- This study presents the time-series analysis, frequency distributions, coefficients of determination and performance matrices for water quality predictability of distinctive Patalganga and Bhatsa rivers by processing eight water quality parameters database of recent two decades (2003 to 2023). The databases of water temperature, Dissolved Oxygen (DOs), Biochemical Oxygen Demand (BOD), pH, conductivity, nitrate + nitrite, fecal coliform and total coliform executions with General Regression and Extreme Gradient Boosting (XGBoost) modelling systems resulted in frequency distributions, coefficients of determinations, performance matrices with Root Mean Squared Error (RMSE), Mean Squared Error (MSE) and Mean Absolute Error (MAE). The results are discussed in a comparative analysis of time-elapsed evolutions with trend-line fits representing the variabilities in water quality over the period of recent 20 years. The BOD and total coliform shows better correlation with coefficient of 0.6 for Bhatsa and a very good correlation of 0.72 between fecal coliform and total coliform for Patalganga. Also, coefficients of determinations of 0.5> recommends fecal coliform, total coliform, BOD, conductivity, Nitrate + Nitrite best water quality predictors and XGBoost performers (RMSE, MSE, or MAE) of <1 supports the same. It means XGBoost model's average prediction error is very low (<1), which seems to be a sign of a good fit, especially the target variable's scale supports such smaller errors. The water quality data-driven analytical results of this case study corroborates the AI-ML forecasting system to be viewed as the balancing approach towards ongoing selective river water samplings with chemical analysis. This system can be the proactive decision-making tool in river water pollution management.