ML-Assisted Hybrid Lightweight Cryptographic Framework for Secure IoT Intrusion Detection

This paper presents an adaptive hybrid lightweight cryptographic framework designed for resource-constrained Internet of Things environments. Conventional cryptographic mechanisms such as the Advanced Encryption Standard provide strong security but incur significant computational overhead, while lightweight ciphers such as SPECK offer efficiency at the cost of reduced security strength. To address this trade-off, we propose a hybrid model integrating AES-128 and SPECK-128/128, enhanced with a machine-learning-based decision module capable of selecting the most suitable algorithm at runtime based on system conditions and threat indicators. The system is implemented using two ESP8266 nodes configured as a real-time intrusion detection network where Node A performs IR-based sensing and encryption, and Node B performs decryption, anomaly detection, and defensive actuation via relay and buzzer. Attack scenarios including replay attacks, tampered ciphertext injection, and flooding are simulated to generate training data. A Decision Tree classifier is trained on features such as encryption time, free heap memory, message frequency, IR state, and plaintext validity. Experimental results demonstrate that the ML-assisted hybrid approach improves detection accuracy and dynamically balances efficiency and security. The framework provides a deployable and cost-effective solution for intrusion-resilient IoT security.