Water quality prediction using machine learning algorithms

Water, which is an essential for the continuity of life, a healthy life and one of the most important cornerstones of development, is losing its usability due to rapid industrialization and the mixing of various pollutants into the ecosystem. The need for clean water resources is increasing day by day for the continuation of a sustainable life and strong economies. Although there are many water resources on earth, clean drinking water resources are limited in many parts of the world and increasing clean water resources is possible by reducing water pollution. In parallel with the increase in industrialization, water quality has deteriorated in almost every country, especially in developed countries. In some countries, water scarcity presents itself as a major problem. In order to increase the amount of water suitable for the required use, it is necessary to constantly monitor the water quality as well as reduce the pollutants released into the ecosystem. Rivers are the main freshwater resources and the solution to the problem for sustainable water management lies in the solution at the source. Surface waters, which are exposed to many pollutants and pollution, are also used as natural discharge areas. Since water quality is of critical importance for the health of people, the sustainability of ecosystems and the welfare of societies, it is practical to use water quality indices instead of all water quality characteristics in monitoring surface water quality. This study focused on predicting water quality with machine learning algorithms. Since the complexity and diversity of factors affecting water quality are both difficult and time-consuming to detect with traditional analytical methods, the use of machine learning algorithms has been used as an effective option in predicting and monitoring water quality. The process was performed first automatically and then with hyperparameter adjustment using machine learning algorithms. Determination of water quality was achieved with the use of a minimum number of independent variables, and an approach that enables water quality monitoring was proposed. Water quality index was estimated using aluminium, ammonium, iron, calcium, chloride, manganese, sulphate, pH, colour, turbidity variables for water quality monitoring taken from the South Australian Government database. Estimation of water quality Gaussian Naive Bayes, K-nearest neighbor (KNN), support vector (Support Vector), artificial neural networks (Artificial Neural Network), decision trees (CART), random forest (Random Forests), gradient boosting (Gradient Boosting Machines), Category Boosting CatBoost and logistic regression models were used.