A Supervised Machine Learning Approach for Analysis and Prediction of Water Quality

Abstract

The need to constantly monitor water quality is paramount as water pollution increases exponentially. In our study, data published by King County, Washington, USA, has been used to compare the performance of several algorithms which were used in predicting water quality index (WQI) and classifying water quality classification (WQC). Multilayer perceptron (MLP) performs most efficiently in both tasks with an RMSE of 0.76, R2 of 0.99, and classification accuracy of 97.1%. Feature analysis shows that phosphorus and ammonia nitrogen have considerable influence on predicting WQC and WQI, respectively, even though they were not directly used in the calculation of WQI. Sensitivity analysis done on the MLP model further shows that after removing the most important feature of turbidity as an input parameter, the model had an RMSE of 12.86 and an R2 of 0.84, respectively. It can be considered as one of the most significant parameters since it affects the WQI drastically.