TinyML-based intrusion detection systems for sustainable and energy-constrained IoT devices

• TinyML for IoT Security : Explores using Tiny Machine Learning (TinyML) for lightweight, energy-efficient intrusion detection on low-power IoT devices. • GraBoost-AAGT Model : Proposes a novel Gradient Boosting-based Adaptive Artificial Gorilla Troops (GraBoost-AAGT) model, combining AAGTO and XGBoost for efficient on-device threat detection. • Optimized for Energy Efficiency : Designed for ultra-low power consumption, the model maintains high detection accuracy while minimizing CPU usage and energy overhead. • Advanced Preprocessing : Utilizes categorical encoding, Z-score normalization, and Fast Fourier Transform (FFT) to extract relevant features for high-performing intrusion detection. • Real-Time Performance : Achieves real-time detection of threats like denial-of-service and probing attacks, with a detection accuracy of 99.50% and F1-score of 99.45%. • Sustainable IoT Security : Demonstrates a sustainable approach to IoT security, enabling secure, scalable, and energy-conscious deployment in environmentally sensitive and resource-constrained settings. With the exponential growth of the Internet of Things (IoT), ensuring real-time security in energy-constrained and environmentally sensitive environments has become a significant challenge. Traditional Intrusion Detection Systems (IDS), though effective, are often resource-intensive and unsuitable for deployment on low-power edge devices. This research explores the potential of Tiny Machine Learning (TinyML) to provide lightweight, on-device intelligence for threat detection. Research proposes a Gradient Boosting Based Adaptive Artificial Gorilla Troops (GraBoost-AAGT), which combines Adaptive Artificial Gorilla Troops Optimization (AAGTO) and Extreme Gradient Boosting (XGBoost), optimized for ultra-low power consumption and memory efficiency. Preprocessing involves categorical encoding and Z-score normalization, followed by Fast Fourier Transform (FFT) to extract relevant frequency-domain features. Specifically designed for embedded IoT systems, this model, referred to as GraBoost-AAGT, supports real-time detection of cyber threats, such as denial-of-service and probing attacks, while maintaining minimal central processing unit (CPU) and energy overhead. To evaluate performance, network traffic data encompassing both normal and malicious activities is collected using IoT intrusion detection dataset, traffic simulators, and replay tools under controlled attack and non-attack scenarios. Results demonstrate that GraBoost-AAGT deployed via TinyML achieves a high detection accuracy (99.50%), precision (99.8%), recall (99.12%), F1-score (99.45%), specificity (99.81%), and computation time (4.5s), reduced energy consumption, and lower CPU usage compared to traditional edge-based ML models. These findings confirm that TinyML offers a viable and sustainable solution for deploying intelligent threat detection in smart and green IoT ecosystems. This research provides a foundational step toward secure, scalable, and environmentally responsible IoT security frameworks.