C-AAE: Compressively Anonymizing Autoencoders for Privacy-Preserving Activity Recognition in Healthcare Sensor Streams

2025-07-23

AI Summary

To address user identity leakage risks in wearable sensor-based activity recognition, this paper proposes the Compressed Anonymous Autoencoder (CAAE) framework, jointly optimizing privacy preservation and recognition utility. Methodologically, CAAE integrates an Anonymous Autoencoder (AAE) to learn de-identified latent representations and incorporates Adaptive Differential Pulse Code Modulation (ADPCM) to perform low-bitrate differential encoding on latent sequences—thereby compressing data volume while simultaneously suppressing residual identity features. Experiments on MotionSense and PAMAP2 datasets demonstrate that CAAE reduces user re-identification F1 scores by 10–15 percentage points, degrades activity recognition F1 by ≤5 percentage points, and achieves a 75% data compression rate. This work is the first to co-design anonymous representation learning with efficient temporal encoding, enabling high-fidelity activity recognition alongside strong privacy guarantees.

Technology Category

• Machine Learning: Privacy
• Planning, Routing, and Scheduling: Activity and Plan Recognition
• Computer Vision: Video Understanding & Activity Analysis

Application Category

• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Security and Privacy: Large-scale security measurements

Abstract

Wearable accelerometers and gyroscopes encode fine-grained behavioural signatures that can be exploited to re-identify users, making privacy protection essential for healthcare applications. We introduce C-AAE, a compressive anonymizing autoencoder that marries an Anonymizing AutoEncoder (AAE) with Adaptive Differential Pulse-Code Modulation (ADPCM). The AAE first projects raw sensor windows into a latent space that retains activity-relevant features while suppressing identity cues. ADPCM then differentially encodes this latent stream, further masking residual identity information and shrinking the bitrate. Experiments on the MotionSense and PAMAP2 datasets show that C-AAE cuts user re-identification F1 scores by 10-15 percentage points relative to AAE alone, while keeping activity-recognition F1 within 5 percentage points of the unprotected baseline. ADPCM also reduces data volume by roughly 75 %, easing transmission and storage overheads. These results demonstrate that C-AAE offers a practical route to balancing privacy and utility in continuous, sensor-based activity recognition for healthcare.

Problem

• Protects user privacy in healthcare sensor data
• Reduces identity leakage while preserving activity recognition
• Compresses data to lower transmission and storage costs

Innovation

• Anonymizing AutoEncoder for privacy protection
• ADPCM for differential encoding compression
• Combined C-AAE balances privacy and utility