Flowing Secure: Proactive AI for Cyberattack Detection in Water Systems

Abstract

Recent rapid advances in digital technologies have triggered a significant digitalization process in critical infrastructure sectors such as water and energy. With this transformation, not only operational efficiency but also how physical systems integrate with their digital counterparts has become a central issue. While this convergence undoubtedly makes systems more efficient, it also creates serious challenges in terms of cybersecurity. The proliferation of Internet of Things (IoT) devices, in particular, has not only exponentially increased the number of connections but also created new and expanded threat surfaces. Every device can become a potential vulnerability. In such an environment, classical security measures are often insufficient. At this point, artificial intelligence-based approaches provide significant advantages. In this study, a model consisting of feature selection, data balancing, and classification processes is proposed for the detection of cyber attacks against SCADA systems. In building the model, meaningful features are selected using the Boruta algorithm, class imbalance is eliminated using the SMOTEENN method, and a GRU-based deep learning structure is used. The proposed system; It has been tested on datasets obtained from critical water networks such as WADI, SWaT, and BATADAL. Accuracy values of 99.85% were achieved for the WADI dataset, 97.78% for the SWAT dataset, and 100% for the BATADAL dataset. The developed deep learning model stands out as an effective and scalable defense mechanism not only against current threats but also against future attacks. © 2025 IEEE.