A Lightweight AI-Framework using PUFs and Chaos-based Encryption for IoT Devices

Abstract

Internet of Things (IoT) devices in critical domains has been mentioned and as a result, it has caused intense pressure on the hardware security. The conventional cryptographic protocols impose a limiting computation and memory overhead on sparse edge devices. The given paper will be an integrated adaptive security architecture, which uses Physically Unclonable Functions (PUFs) to generate hardware-grounded keys, chaos-based Logistic Map encryption to secure lightweight payloads, HMAC-SHA256 to track nonces to ensure integrity and replayability, and a Random Forest regression system to sustain the well-being of devices with the help of AI. The system was functionally evaluated on an ESP32 microcontroller against six attack scenarios, including replay, delayed replay, payload tampering, forged device identity, MAC forgery, and unknown-device injection. All tested attack cases were successfully detected and rejected, with no false acceptance observed during the evaluation. The results of the experiment confirm the following: the throughput of 187.6 KB/s with the model R 2 equal to 0.9312 and the MAE at the level of 1.47 health-score points. The architecture offers both multi-layered and energy-efficient security to massive IoT-based applications.