A COMPREHENSIVE REVIEW ON EDGE AI-BASED WATER QUALITY MONITORING, PREDICTION AND ALERT SYSTEM

Abstract

Access to clean drinking water is one of the most basic requirements for human health, yet a large part of the world still relies on water testing methods that are slow, expensive, and reactive rather than preventive. Conventional laboratory-based water testing can take two to three days to deliver results and can cost anywhere between 3000 and 4000 per sample, which makes frequent, widespread testing impossible for many households, schools, and small municipalities. Over the past decade, Internet of Things (IoT) based monitoring systems have tried to close this gap by placing sensors directly into water bodies and streaming the readings to cloud servers for processing and visualization. While this is a major step forward compared to manual testing, it introduces a new weakness: the entire safety function of the system now depends on a stable internet connection. Any network outage, server downtime, or connectivity delay can silently disable the alerting function exactly when it is needed most, such as during floods or infrastructure failures. This paper presents a detailed review of Edge Artificial Intelligence (Edge AI) as a solution to this problem. In an Edge AI system, the intelligence needed to interpret sensor data is moved away from remote cloud servers and placed directly onto the low-cost microcontroller attached to the sensors. We review fifteen recent research works published between 2024 and 2026 covering cloud-based sensor nodes, long-range wireless networks, solar-powered systems, AI based forecasting, specialized chemical sensors, and offline handheld testers, and we identify a consistent gap: no existing system combines multi-parameter sensing, on-device Water Quality Index (WQI) computation, instant local alerting, remote dashboard access, and low cost into a single practical design. Building on this gap, we propose a simple, low-cost Edge AI architecture using the ESP32 microcontroller that reads pH, Total Dissolved Solids (TDS), turbidity, and temperature sensors, calculates the WQI directly on the chip, and activates a local LED and buzzer alarm within milliseconds if the water is found unsafe, entirely independent of internet availability. Data is also optionally forwarded to an online dashboard whenever a connection is present. This approach is estimated to reduce alert latency by more than 95% and hardware cost by up to 60% compared to cloud-dependent or single-board-computer-based systems, while remaining simple enough to build and deploy in homes, schools, and rural communities. The paper closes with future research directions, including Tiny ML-based anomaly detection, solar-powered autonomous operation, and long-range LoRa communication.