WATER QUALITY INTELLIGENCE MONITORING NETWORK (WATER QI)

Abstract

Water is an essential resource for the survival of humans, animals, and plants. However, despite its critical importance, clean and safe water is not always readily available for drinking, household needs, or industrial applications. A variety of factors—including industrial activities, mining operations, pollution, and natural environmental processes—can influence water quality by introducing contaminants or altering its physicochemical properties. These changes directly affect the suitability of water for human consumption or general usage. To regulate this, the World Health Organization (WHO) provides guidelines that specify acceptable threshold levels for various water quality parameters used for drinking and irrigation.Metrics such as the Water Quality Index (WQI) and the Irrigation Water Quality Index (IWQI) consolidate these parameters into single numerical indicators to evaluate overall water quality. However, the traditional process of collecting samples from diverse sources, transporting them under controlled conditions, measuring multiple parameters, and comparing results with established standards is labor-intensive, time-consuming, and operationally challenging.To address these limitations, this study introduces a real-time monitoring framework that integrates a distributed network architecture with machine learning (ML) techniques to automatically assess the suitability of water samples for drinking and irrigation. The proposed network utilizes LoRa communication technology and is designed with consideration of land topology. Simulation results generated using Radio Mobile identified a partial mesh topology as the most effective structure for data transmission.Due to the lack of comprehensive public datasets for drinking and irrigation water quality, custom datasets were created to train ML models. Three algorithms—Random Forest (RF), Logistic Regression (LR), and Support Vector Machine (SVM)—were evaluated for water classification tasks. Experimental findings revealed that LR achieved the highest accuracy for drinking water classification, while SVM proved most effective for irrigation water assessment. Additionally, recursive feature elimination was applied in conjunction with these models to identify the most influential water quality parameters contributing to classification performance.