Supervisory Control an‎d Data Acquisition (SCADA) systems play a critical role in managing an‎d monitoring industrial infrastructures such as electricity, oil an‎d gas, water treatment, an‎d manufacturing. With the increasing integration of SCADA networks with IT systems an‎d the Internet, these critical infrastructures have become more vulnerable to sophisticated cyber threats, particularly Distributed Denial of Service (DDoS) attacks. Such attacks can disrupt industrial operations, cause significant financial losses, an‎d threaten public safety. Traditional intrusion detection systems (IDS), especially signature-based approaches, often fail to detect zero-day an‎d complex attacks, highlighting the need for more intelligent an‎d adaptive detection mechanisms. This research examines the enhancement of DDoS attack detection in SCADA environments through the integration of feature selec‎tion techniques an‎d a hybrid deep learning model based on Long Short-Term Memory (LSTM) an‎d Dense layers. Feature selec‎tion is employed to reduce data dimensionality, eliminate redundant an‎d irrelevant attributes, an‎d improve model efficiency an‎d generalization. The LSTM component captures temporal an‎d sequential patterns in network traffic, while Dense layers model complex non-linear relationships among selec‎ted features, improving overall classification performance. By combining intelligent feature selec‎tion with deep learning, the proposed approach aims to achieve higher detection accuracy, reduced false alarm rates, an‎d improved robustness against evolving cyber threats. This study contributes to strengthening the cybersecurity of SCADA systems an‎d provides a framework for developing advanced, reliable, an‎d real-time intrusion detection solutions for critical industrial infrastructures.

2/18/2026 12:00:00 AM

89808

1404/12/05

Enhanced DDoS Detection in SCADA Systems Using Feature selec‎tion an‎d LSTM-Dense Hybrid Models

DDoS , SCADA , Random Forest , LSTM-Dense Hybrid