Neural network method of an analysis and forecasting water quality parameters to ensure public health safety

Introduction. Water quality is a critical factor in public health and sustainable environmental development. With increasing anthropogenic pressure and climate change, traditional methods of water quality monitoring are not sufficiently effective, requiring the innovative approaches based on artificial intelligence.Objective. To develop and test a hybrid neural network model for accurate prediction water quality indicators.Materials and methods. The study used a set of software tools on the Python platform with the use of machine learning libraries (TensorFlow Keras, Scikit-learn, Pandas). Preliminary data processing was performed using iForest and Lagrange interpolation methods. An original hybrid architecture was developed, combining convolutional neural networks and bidirectional recurrent layers (BiGRU). The model was trained on real water quality monitoring data using fifty epochs and 120 time intervals. The accuracy of the forecasts was evaluated using four metrics (RMSE, MAE, MAPE, R2) and then compared with the traditional ARIMA and SMA methods.Results. The hybrid neural network model demonstrated unprecedented forecasting accuracy with a coefficient of determination R2>0.995 for key water quality parameters. The mean prediction error (RMSE) was 0.0309, which is 44.5% better than the results of the closest analogue and 32.2 times better than classical methods. The time required to generate a forecast for the next day does not exceed 0.83 seconds. The model works effectively in conditions of stable indicators, identifying key factors affecting water quality.Limitations. The analysis was conducted on data for the period 2014–2022, which limits the assessment of long-term trends and extreme events. Model performance decreases when there are more than 5% anomalous values ​​and gaps in the original data without pre-processing.Conclusion. The developed hybrid neural network model is an effective tool for operational forecasting and control of water resources quality. Its application opens up opportunities for creating early warning systems for potential threats to public health, optimizing water treatment processes, and developing preventive measures to protect aquatic ecosystems. The high accuracy and speed of the model ensure its practical applicability in real-world drinking water source monitoring systems.Compliance with ethical standards. This study does not require the conclusion of a biomedical ethics committee or other documents.Contribution: Shiryayeva M.A. — research concept and design, data collection and processing, text writing; Pushkareva M.V. — data collection and processing, text writing. All authors are responsible for the integrity of all parts of the manuscript and approval of the manuscript final version.Conflict of interest. The authors declare no conflict of interest.Funding. The study had no sponsorship.Received: November 21, 2025 / Revised: December 9, 2025 / Accepted: December 19, 2025 / Published: February 10, 2026