A Hybrid Machine Learning Framework for Water Quality Index Prediction Using Feature-Based Neural Network Initialization

Accurate prediction of the Water Quality Index (WQI) is crucial for safeguarding public health and managing freshwater resources effectively. Traditional machine learning models often face challenges due to arbitrary weight initialization and the limited use of ensemble learning, leading to unstable predictions and reduced interpretability. In this study, we introduce a novel hybrid machine learning framework designed to address these issues by combining feature-informed neural network initialization with the power of ensemble methods such as XGBoost, Random Forest, Stacking, Hist-Gradient Boosting, Deep Residual Network and Voting Regressor. The neural network weights are initialized using feature significance scores derived from SHapley Additive exPlanations (SHAP), which provide a more informed and interpretable starting point. Predictions are then refined iteratively using the ensemble methods to enhance model performance and robustness. This framework leverages the strengths of both feature-based initialization and ensemble learning to produce more reliable and interpretable predictions of the WQI, paving the way for better water quality management and resource protection.

Keywords: Water Quality Index (WQI), Hybrid Machine Learning Framework, Feature-Informed Neural Network Initialization, XGBoost, Random Forest, Stacking, Voting Regressor, Shapley Additive explanations (SHAP), Ensemble Learning, Interpretability, Water Resource Management