Energy-Efficient AI Framework Revolutionizes Cybersecurity for Smart and IoT Infrastructures

In a cross-continental partnership, researchers from the University of South Florida, SRM Institute of Science and Technology, KG Reddy College of Engineering and Technology, and Sathyabama Institute of Science and Technology have pioneered an intelligent, energy-efficient cybersecurity model aimed at protecting the rapidly expanding universe of connected devices. Published in Results in Engineering (Elsevier, 2025), their paper, "AI-driven energy-efficient cybersecurity frameworks for sustainable digital infrastructures", introduces the Dynamic Beetle Antennae Search–Mutated Adaptive Weighted Random Forest Tree (DynBAS-AWRF Tree). The model is designed to tackle one of today's most pressing challenges: how to defend complex IoT and edge computing systems from cyberattacks without exhausting energy resources.

Balancing Power, Precision, and Sustainability

The study begins by highlighting a crucial dilemma. As billions of IoT devices enable smart cities, healthcare networks, and industrial automation, the systems that safeguard them are becoming energy-intensive and unsustainable. Traditional intrusion detection models, though effective, are often too heavy for power-constrained devices. The researchers emphasize that future cybersecurity must balance detection precision with energy economy, enabling systems to remain both secure and sustainable. Their solution, a hybrid AI model that learns adaptively and operates efficiently, reflects this dual focus. It uses minimal computational resources while maintaining high-speed, high-accuracy detection across diverse digital environments.

The Science Behind the DynBAS-AWRF Tree

At the heart of the framework lies an elegant combination of two AI mechanisms. The Dynamic Beetle Antennae Search (DynBAS) algorithm, inspired by the sensory movement of beetles, intelligently searches for optimal data features and fine-tunes model parameters. It employs adaptive step control and multi-directional search strategies to avoid getting stuck in local optima, ensuring fast and accurate optimization. Alongside it, the Adaptive Weighted Random Forest (AWRF) algorithm assigns performance-based weights to decision trees, adjusting dynamically to handle unbalanced datasets where certain attacks are rare. This combination allows the system to identify both common and uncommon cyber threats quickly while using far less energy than traditional models.

The researchers trained their model on the TON-IoT dataset, a benchmark containing telemetry data from IoT devices operating under normal and attack conditions. To improve learning efficiency, they employed Min–Max normalization, one-hot encoding, and Linear Discriminant Analysis (LDA) for dimensionality reduction. This streamlined data pipeline reduced redundancy and enabled real-time decision-making even on small, low-power devices.

High Accuracy, Low Energy, Real Impact

The DynBAS-AWRF Tree produced remarkable outcomes. It achieved 96% detection accuracy, 85% energy efficiency, and an AUC score of 0.90, while cutting feature usage by 45% and inference time to just 0.25 milliseconds. The model demonstrated strong performance across attack types, 95% for denial-of-service, 90% for exploit-based intrusions, and over 85% for reconnaissance attempts. Figures in the paper show balanced classification results with minimal false alarms and impressive adaptability to real-world conditions.

Equally important, the system becomes more efficient as it scales. Increasing the number of processing nodes from 100 to 1000 raised energy efficiency from 50% to 85% and reduced execution time from 0.40 to 0.25 milliseconds. This scalability positions the framework as a leading contender for next-generation cybersecurity in resource-limited environments such as industrial IoT, healthcare monitoring, and smart grids.

Setting a New Standard for Green AI

A comparative analysis showed that DynBAS-AWRF Tree outperformed existing "green AI" models, including Karamchand's 2025 framework. In financial fraud detection, it reduced energy consumption from 180 to 120 kilowatt-hours; in healthcare monitoring, from 220 to 180 kWh; and in IoT protection, from 200 to 130 kWh. Across sectors, cloud security, finance, and government infrastructure, the model consistently maintained top-level accuracy while slashing energy use.

Beyond its technical performance, the framework also aligns with global sustainability and data governance goals such as GDPR, NIST, and the EU Green Digital Charter. By performing localized, on-device analytics, it minimizes data transfer, latency, and power consumption, making cybersecurity not just stronger but also environmentally responsible. The authors position this as a step toward "green cybersecurity," where artificial intelligence serves both technological innovation and ecological balance.

The Road Ahead: Toward Sustainable Digital Defense

The study concludes that DynBAS-AWRF Tree represents a major advance in sustainable AI-driven security. By combining adaptive intelligence with low-energy operation, it enables scalable protection for emerging digital infrastructures. However, the researchers acknowledge challenges, such as the dependence on high-quality labeled data and deployment constraints in small-edge devices. Future work, they suggest, will explore federated learning, renewable-powered data centers, and hardware–software co-design to enhance resilience and scalability.

In fewer than a second, this system can detect and neutralize cyber threats while consuming nearly half the power of conventional models. As digital ecosystems grow ever more interconnected, the DynBAS-AWRF Tree offers a visionary path forward, one where cybersecurity and sustainability evolve hand in hand, shaping a future in which intelligent defense is as energy-aware as it is powerful.