"ML-Enhanced ChaCha20 for Constrained-Hardware IoMT Encryption: A Reproducible 3.7\u00d7 Energy-Memory Gain"

"Medical Internet of Things (IoT) devices face critical constraints in energy consumption and memory footprint while requiring real-time encryption for sensitive patient data. This paper presents a reproducible, energy-memory optimized encryption framework that integrates machine learning (ML) for adaptive parameter selection with hardware-aware optimizations. The proposed system is validated on 30 real electrocardiogram (ECG) recordings from the MIT-BIH Arrhythmia Database, comprising 54,000 samples at 360 Hz. Standard ChaCha20 achieves 0.742 ms encryption time with 23.7 \u03bcJ energy consumption and a 2.0 KB memory footprint for 1-second ECG data (1440 bytes). Compared to AES-256, ChaCha20 provides 3.7\u00d7 lower energy consumption, 2.4\u00d7 faster encryption, 42.9% memory reduction, and 3.7\u00d7 longer battery life (3,247 days versus 886 days on a 500 mAh battery). The ML-enhanced variant further improves speed to 0.081 ms, an 8.0% improvement over standard ChaCha20. Critically, NIST SP 800-22 statistical testing confirms that all encryption outputs pass all 9 tests (100% pass rate), demonstrating that the encrypted data is statistically indistinguishable from truly random sequences. All experiments were conducted on an Intel Core i3 processor; the effective memory footprint remained under 15 KB\u2014well within the 256 KB constraint of typical medical IoT devices. The complete code and dataset are publicly available for reproducibility"