A deep learning-based IoT malware detection approach for electric vehicle charging stations

Area of Science:

• Cybersecurity
• Internet of Things (IoT) Security
• Machine Learning for Security

Background:

• Electric Vehicle (EV) charging stations increasingly rely on Internet of Things (IoT) devices, necessitating robust security measures.
• Existing IoT malware detection methods struggle with cross-architecture compatibility, limited feature extraction, and simplistic data fusion techniques.

Purpose of the Study:

• To develop a unified and adaptable deep learning approach for detecting IoT malware in EV charging station environments.
• To overcome the limitations of existing methods by enabling cross-architecture analysis and comprehensive feature representation.

Main Methods:

• Utilized a decompilation tool to convert opcodes into a unified intermediate representation (PCode) for cross-architecture compatibility.
• Integrated global structural, statistical, and semantic features, processing each with dedicated deep learning algorithms.
• Implemented a multimodal feature fusion model with dynamic weighting and a multi-layer encoder for deep data integration.

Main Results:

• The proposed deep learning approach demonstrated superior performance compared to existing methods on public IoT malware datasets.
• Achieved a 1.37% improvement in F1 score, validating the effectiveness of the unified and dynamic fusion strategy.

Conclusions:

• The developed method offers a significant advancement in IoT malware detection for critical infrastructure like EV charging stations.
• The approach provides a more adaptable, comprehensive, and effective solution for securing heterogeneous IoT ecosystems.