PREDICTION OF WATER QUALITY INDEX WITH MACHINE LEARNING METHODS

Abstract

Water, which is the most important part of life and development, loses its usability due to increasing pollution sources. Clean water resources must be increased in order for living things to survive in a healthy way and for economies to grow. However, water resources are limited and the increase in clean water resources can be possible by reducing water pollution. Today, water quality deteriorates in almost every country. In some countries, this problem manifests itself as water scarcity. Water quality should be monitored regularly and continuously in order to increase the amount of usable water. It is important to solve the problem at the source for sustainable water management. Rivers are the main sources of fresh water. However, since they are used as a natural discharge area, they are exposed to a lot of pollution. It is practical to use water quality indices instead of all water quality characteristics in the monitoring of surface water quality. In this study, the performance of machine learning methods for monitoring surface water quality with sensors was compared. Since sensors for water quality characteristics are expensive, the use of a minimum number of parameters is intended. It has been tried to present an approach that enables on-site water quality monitoring by reducing the variables in the water quality index and using fewer sensors. For the water quality monitoring of the Porsuk river, the water quality index was estimated using dissolved oxygen, fecal coliform, turbidity and total solid variables. By using the sensors of these variables and the XGBoost method, it has been determined that the water quality index can be monitored effectively and continuously with less cost.