Bayesian optimised Ensemble Learning based verifier-like Attestation Framework designed for IoT Network Security aiding in multiclass malware detection.
Attestation entails presenting verifiable evidence to an evaluator to substantiate claims regarding a target’s characteristics, ensuring that the firmware and configuration are reliable, verifying that the hardware is genuine. This paper proposes attestation solutions based on optimized ensemble learning and neural networks to bridge the existing gaps, enhancing security in the device lifecycle.
Ensemble models with bayesian hyperparameter tuning:
- Maximum Soft Voting Ensemble
- Stack Ensemble
- ML models were trained on a feature-engineered realistic cyber security dataset that covers Ransomware, Man in the middle and SQL Injection attacks, to name a few out of 14 other attacks detected in the IoT and Industrial IoT Perception Layer, gathered from 10 different types of IoT devices.
- Based on the data analysis of testbed and evaluating the performance of the proposed approaches, the framework promises an accuracy of 93% with 92% precision for detection of compromised IoT device and attack type.
- Stacking ensemble’s accuracy outperformed CNN model’s accuracy with early stopping callback in terms of multi-label classification which resulted in an accuracy of 81% with 80% precision.