Artificial IoT water quality analysis system: A machine learning-driven approach for small-scale real-time evaluation

Traditional water monitoring pipelines suffer from structural inefficiencies that cause significant analytical time lags, preventing immediate intervention during sudden water quality fluctuations and leading to severe economic losses like mass fish mortality. To address this, we introduce WaterMirror and WQSurrogateModels, which are integrated IoT and Machine Learning (ML) framework designed to automate expertise-intensive monitoring. Using real-world historical data, the system performs surrogate regression of the WQI 5 index across five critical parameters: Dissolved Oxygen (DO), Biological Oxygen Demand (BOD), Ammonia Nitrogen (NH3-N), Electrical Conductivity (EC), and Suspended Solids (SS). A comparative analysis of six ML models demonstrates that tree-based ensemble methods significantly outperform traditional architectures. Specifically, the experiment demonstrated that even in the absence of the NH3-N indicator, the monitoring system maintained high functionality of warning tasks for alert generation. Furthermore, stress tests confirmed the model's sensitivity to sudden spikes in organic pollution. Released under the Apache License 2.0, WaterMirror and WQSurrogateModels offer a scalable, open-source solution for real-time, high-precision environmental assessment.