Enhancement of water quality index prediction using support vector machine with sensitivity analysis

Abstract

For more than 25 years, the Department of Environment (DOE) of Malaysia has implemented a water quality index (WQI) that uses six key water quality parameters: dissolved oxygen (DO), biochemical oxygen demand (BOD), chemical oxygen demand (COD), pH, ammoniacal nitrogen (AN), and suspended solids (SS). Water quality analysis is an essential component of water resources management that must be properly managed to prevent ecological damage from pollution and to ensure compliance with environmental regulations. This increases the need to define an efficient method for WQI analysis. One of the major challenges with the current calculation of the WQI is that it requires a series of sub-index calculations that are time consuming, complex, and prone to error. In addition, the WQI cannot be calculated if one or more water quality parameters are missing. In this study, the optimization method of WQI was developed to address the complexity of the current process. The potential of data-driven modeling, i.e., Support Vector Machine (SVM) based on Nu-Radial basis function with 10-fold cross-validation, was developed and explored to improve the prediction of WQI in Langat watershed. A thorough sensitivity analysis under six scenarios was also conducted to determine the efficiency of the model in WQI prediction. In the first scenario, the model SVM-WQI showed exceptional ability to replicate the DOE-WQI and obtained statistical results at a very high level (correlation coefficient, r > 0.95, Nash Sutcliffe efficiency, NSE >0.88, Willmott’s index of agreement, WI > 0.96). In the second scenario, the modeling process showed that the WQI can be estimated without any of the six parameters. It can be seen that the parameter DO is the most important factor in determining the WQI. The pH is the factor that affects the WQI the least. Moreover, scenarios three to six show the efficiency of the model in terms of time and cost by minimizing the number of variables in the input combination of the model (r > 0.6, NSE >0.5 (good), WI > 0.7 (very good)). In summary, the model will greatly improve and accelerate data-driven decision making in water quality management by making data more accessible and attractive without human intervention.