Aprendizado de máquina supervisionado e não supervisionado aplicado ao monitoramento da qualidade de águas superficiais e subterrâneas na bacia do Rio Mogi Guaçu (UGRHI-9)

Abstract

Environmental monitoring of surface and groundwater is fundamental to preserving water quality and the health of different ecosystems. Considering the difficulties in the monitoring process, which is highly time-consuming and costly, this work aims to integrate data science tools, particularly machine learning (ML) methods, into natural water quality monitoring based on historical data from the São Paulo State Environmental Company (CETESB). By means of supervised learning (Multiple Linear Regression and Random Forest), the models were trained to predict the IQA (Water Quality Index), obtaining excellent test results for both models (Regression with R² of 0.97 and RMSE of 2.37 and Random Forest with R² of 0.92 and RMSE of 0.06), thus allowing the prediction of surface water quality in the future. Similarly, unsupervised learning (PCA, K-Means and DBSCAN) was applied in order to detect patterns, analyze correlations and reduce the dimensionality of the variables, which could have generated better results if more data had been available. In general, however, the algorithms provided satisfactory results that could reduce the time and costs involved in monitoring water quality.