Title: Design And Development Of Machine-Learning Based Anomaly Detection Framework For Iot Water Quality Data
Authors: Isaya M. Chogoro, Halima R. Masiki, Gasto Ndaga, Lucas Misana, Maombi Mwasumbi, Johnson Abihudi, Janeth Kaneno, Jackline Donald
Volume: 10
Issue: 5
Pages: 223-231
Publication Date: 2026/05/28
Abstract:
Access to safe drinking water remains a critical challenge in low-resource communities, where traditional water quality monitoring is slow, expensive, and reactive. While Internet of Things (IoT) sensors enable continuous data collection, most existing systems rely on simple threshold-based alerts that fail to detect complex, multivariate contamination events. This paper reviews the state of the art in machine learning (ML)-based anomaly detection for IoT water quality data, focusing on unsupervised methods suitable for environments where labelled contamination events are rare. Two prominent algorithms-Isolation Forest (tree-based) and LSTM autoencoder (deep learning)-are examined in detail. The review synthesizes empirical studies on IoT architectures, data preprocessing, and model evaluation metrics (precision, recall, F1-score). Key challenges including sensor calibration, data noise, and deployment in low-resource settings are discussed. The paper concludes that hybrid approaches and rigorous comparison of algorithms on live sensor streams remain important research directions for early contamination warning and public health protection.