WYSIWYG: IoT Device Identification Based on WebUI Login Pages

Area of Science:

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

Background:

• Internet of Things (IoT) devices are increasingly vulnerable to cyber threats due to widespread intelligence and interconnection.
• Effective device identification is crucial for cybersecurity operations like asset management and vulnerability response.
• Current identification methods can be limited, necessitating novel approaches for enhanced security.

Purpose of the Study:

• To propose a novel method for identifying Internet of Things (IoT) devices using their WebUI login pages.
• To develop an ensemble learning model for accurate vendor identification.
• To create an Optical Character Recognition (OCR)-based method for precise device type and model identification.

Main Methods:

• Utilized distinctive vendor-specific characteristics from WebUI login pages as data sources.
• Developed an ensemble learning model combining Convolutional Neural Networks (CNN) and Deep Neural Networks (DNN) for vendor identification.
• Implemented an Optical Character Recognition (OCR) based method for device type and model identification.

Main Results:

• The ensemble model achieved 99.1% accuracy and 99.5% F1-Score for vendor determination.
• Vendor identification accuracy reached 98% with a 98.3% F1-Score.
• The OCR-based method attained 99.46% accuracy in device model identification, surpassing Shodan by 11.39%.

Conclusions:

• The proposed WebUI login page analysis method offers a highly accurate and effective approach to IoT device identification.
• The combination of ensemble learning and OCR significantly enhances the precision of vendor, type, and model identification.
• This method provides a substantial improvement over existing tools like Shodan for IoT device characterization and security.