Prediction of water quality parameters of Nahand River using random forest model optimized with genetic algorithm

This research aims to enhance the accuracy of water quality predictions using machine learning models. The focus was on evaluating the performance of the Random Forest (RF) model and its hybrid version with a Genetic Algorithm (GA-RF) in predicting biochemical oxygen demand (BOD) and dissolved oxygen (DO) in the Nahand River Basin, Iran. The hybrid model was developed using eleven years of daily water quality data from 2013 to 2023, incorporating 11 input variables—including total nitrogen, total phosphate, nitrite, nitrate, phosphate, nephelometric turbidity unit, water temperature, air temperature, electrical conductivity, pH, and flow. Additionally, six scenarios were created using different combinations of these inputs. The models' performance was statistically evaluated through the coefficient of determination (R²), root mean square error (RMSE), Nash-Sutcliffe efficiency (NS), and Willmott’s index of agreement (WI). Results demonstrated that GA-RF consistently outperformed the standalone RF. In BOD prediction, the GA-RF-6 and RF-5 models achieved R² values of 0.563 and 0.548, respectively. For DO prediction, GA-RF-5 and RF-6 yielded R² values of 0.81 and 0.792, respectively. The findings indicate that integrating the Genetic Algorithm with Random Forest can enhance predictive accuracy in water quality assessments, supporting more informed and sustainable water resource management decisions.