Lightweight Machine Learning-Deep Learning Framework for IoT Devices
Otily Toutsop, Tsion M. Yimer, Kevin Kornegay, Flavien Donfack, Marcial Tienteu · 2025
The Internet of Things (IoT) has surpassed the market and evolved rapidly with many applications in different domains, such as healthcare, building automation systems, agriculture, home automation systems, and smart cars. The market growth of IoT devices is projected to reach 25 billion by 2025. The market explosion has created a security risk as many devices do not have adequate built-in security, opening the door for hackers to exploit security vulnerabilities further. IoT ecosystems, including smart cameras, industrial control systems, medical devices, and others, usually operate with minimal memory, power, and processing capabilities, making traditional security deployed on devices such as antivirus software and deep learning-based intrusion detection models sophisticated for resource-constrained devices. IoT device manufacturers often focus on rapid deployment, focusing on cost over robust security. Therefore, many systems have default passwords and unencrypted communications protocols, and they usually update firmware, which third-party hackers can exploit. Recognizing these vulnerabilities, this research explores and proposes novel and secure lightweight machine learning and deep learning models for resource-constrained devices to optimize their security posture. The framework proposed a lightweight algorithm to establish and deploy innovative IoT infrastructure models. Moreover, we propose lightweight machine learning and deep learning models for low-power devices. Our experimental results provided the following: The Lightweight ANN:76%, lightweight BNN: 78%, lightweight LSTM: 76%, lightweight LSTM: 79%, Lightweight FNN:75% and Lightweight XGBoost: 98%.