Detecting Inference Attacks Involving Raw Sensor Data: A Case Study.

Area of Science:

• Computer Science
• Cybersecurity
• Data Mining

Background:

• Home sensors generate data for health and energy insights.
• New algorithms create inference channels, risking personal data breaches via inference attacks involving raw sensor data (IASD).
• Existing detection systems struggle with new inference channel representations and user knowledge.

Purpose of the Study:

• Propose RICE-M for representing inference channels.
• Develop RICE-Sy, an extensible system to detect IASDs.
• Optimize detection complexity and response times for IASD detection.

Main Methods:

• Introduced RICE-M for inference channel representation.
• Developed RICE-Sy system for IASD detection, evaluated on the MHEALTH dataset.
• Implemented conceptual optimizations and partitioning strategies for complexity reduction.
• Proposed H-RICE-SY hybrid architecture for efficient detection.

Main Results:

• IASD detection complexity is initially quadratic, reduced to linear with optimizations.
• Partitioning strategies reduced median detection time by 63%.
• H-RICE-SY maintained detection times under 80 ms for 30% malicious users up to 1.2 million entities.
• Estimated processing of 8.6 million user information entities with 5% malicious knowledge.

Conclusions:

• RICE-M and RICE-Sy effectively detect IASDs.
• Optimizations significantly improve detection efficiency and scalability.
• Hybrid architecture provides efficient online detection within acceptable time limits.