Monitoring Water Quality in Pusu River Using Internet of Things (IoT) and Machine Learning (ML)

Abstract

The availability of clean water, a vital natural resource that supports diverse ecosystems, is increasingly threatened by sediment accumulation which impacts rivers, oceans, and coastal life, which is in line with sustainable development number goal 6 clean water and sanitation. Rapid industrialization and urbanization have intensified these challenges, leading to the degradation of natural water ecosystems and placing an undue strain on water resources. Pollution from sediments and human activities carries harmful contaminants, reduces visibility, disrupts aquatic life, and impairs ecosystem function. Maintaining the health of rivers and other water bodies requires the timely detection of changing conditions and deterioration, which is crucial for implementing effective countermeasures. However, current water quality monitoring methods primarily rely on laboratory tests, which require specialized staff, chemicals, and expertise. These traditional methods are often insufficient for addressing the complex and dynamic issues of water quality. Fortunately, the advent of the Internet of Things (IoT) technology has enabled real-time collection of water quality data. In addition, the application of soft computing technology for water quality assessment offers a more efficient, faster, and environmentally friendly alternative to conventional laboratory-based techniques. In this dissertation, we propose the use of an IoT device to monitor the performance of a water treatment system and collect data on key water quality indicators. Machine learning (ML) tools will be employed to analyze and simulate these data, enabling the prediction of future water quality parameters. The water quality dataset was collected in two stages. During the first iteration, data were gathered using sensors that measured four parameters: pH, turbidity, temperature, and total dissolved solids (TDS). In the subsequent iteration, the dataset was expanded to include a dissolved oxygen sensor in addition to the initial four sensors. The data collection process for turbidity and other water quality parameters involved more than just 879 data points, the data collection process was comprehensive, and the dataset was validated and analyzed with seasonal changes in mind, systematic approach ensured that the water quality parameters data collected were reliable, accurate, and actionable for monitoring water quality in the river. The dataset encompasses samples from three distinct potability classes: potable water sources, free-flowing river water from the Pusu River, and stagnant water from the puddles, and potholes. Nine proven classification algorithms were applied to the datasets, successfully classifying the water quality conditions with up to 98% accuracy. The best-performing model was then deployed and integrated into a graphical user interface (GUI) for rapid water condition testing, thereby facilitating the instantaneous assessment of water quality.