Bootstrap approach for quantifying the uncertainty in modeling of the water quality index using principal component analysis and artificial intelligence

Collecting and analyzing data on surface water across extensive areas is a challenging, time-consuming and expensive. Developing predictive models that offer high accuracy, reliability and require minimal parameters can potentially reduce the time and expense associated with water quality monitoring and management. While most existing studies have focused on estimating point prediction of water quality without approximating the predictive interval (PI) of the estimation, this study aimed to develop a prediction tool to estimate the PI of water quality indexes (WQIs) in the lower Mun river basin. This was achieved by employing principal component analysis (PCA), artificial neural networks (ANN), and bootstrap methods to enhance accuracy, robustness, and reliability with the minimum number of water quality parameters. PCA was initially used to select 4 parameters for the WQI. Subsequently, ANN regression was employed to develop a new WQI to enhance data evaluation efficiency. The testing results of the proposed model revealed its excellent performance compared to other models in terms of accuracy (root mean square error (RMSE) = 0.86, correlation coefficient (R) = 0.993, scatter index (SI) = 0.019, mean absolute error (MAE) = 0.709, and mean bias error (MBE) = −0.003). Additionally, the proposed model incorporated the bootstrap method to quantify uncertainty and create a PI, resulting in a high coverage rate exceeding 95%. By integrating statistical techniques with artificial intelligence and quantifying uncertainty, it is possible to effectively evaluate water quality, provide more accurate and reliable indexes. This study can be an effective tool for decision makers and planners seeking precise data on water quality to develop water resource management strategies.