CLASSIFICATION OF WATER QUALITY BY DIFFERENT ARTIFICIAL INTELLIGENCE ALGORITHMS

The water and the monitoring of its quality is an area of great importance for modern society. Indexes for assessing water quality (WQI) have been developed for many years, serving as easy-to-interpret tools. However, with the increase in data collection and consequently an increase in the complexity of systems, there is a need to use automated and more modern techniques. With the increasing use of machine learning algorithms to optimize processes, this work proposed to evaluate 10 classic machine learning models (LDA – Linear Discriminant Analysis, QDA – Quadratic Discriminant Analysis, LR – Logistic Regression, Perceptron, RC – Ridge Classifier, GNB – Gaussian Naive Bayes, KNN – Knearest Neighbors, SVM – Support Vector Machine, DT – Decision Tree and MLP – Multilayer Perceptron) and 5 ensemble models (ADB - AdaBoost, BAG - Bagging, ET - Extra Tree, GDB - Gradient Boosting and RF – Random Forest). The statistical evaluation of the obtained models was made using the following metrics: balanced accuracy, precision, recall and f1, confusion matrix. Then, the number of variables was reduced to ensure an acceptable classification and the limit of 4 features was reached (tc – thermotolerant coliforms, bod – biochemical oxygen demand, do – dissolved oxygen and tp – total phosphorus). The SVM and GDB techniques stood out as they obtained the best metrics. This work demonstrated the feasibility of using artificial intelligence models to classify water quality, as well as the possibility of predicting this classification with a smaller number of measured variables.