A Genetic Attack Against Machine Learning Classifiers to Steal Biometric Actigraphy Profiles from Health Related Sensor
- PDF / 1,103,444 Bytes
- 11 Pages / 595.224 x 790.955 pts Page_size
- 52 Downloads / 160 Views
IMAGE & SIGNAL PROCESSING
A Genetic Attack Against Machine Learning Classifiers to Steal Biometric Actigraphy Profiles from Health Related Sensor Data Enrique Garcia-Ceja1 · Brice Morin1 · Anton Aguilar-Rivera2 · Michael Alexander Riegler3 Received: 22 October 2019 / Accepted: 20 August 2020 / Published online: 15 September 2020 © The Author(s) 2020
Abstract In this work, we propose the use of a genetic-algorithm-based attack against machine learning classifiers with the aim of ‘stealing’ users’ biometric actigraphy profiles from health related sensor data. The target classification model uses daily actigraphy patterns for user identification. The biometric profiles are modeled as what we call impersonator examples which are generated based solely on the predictions’ confidence score by repeatedly querying the target classifier. We conducted experiments in a black-box setting on a public dataset that contains actigraphy profiles from 55 individuals. The data consists of daily motion patterns recorded with an actigraphy device. These patterns can be used as biometric profiles to identify each individual. Our attack was able to generate examples capable of impersonating a target user with a success rate of 94.5%. Furthermore, we found that the impersonator examples have high transferability to other classifiers trained with the same training set. We also show that the generated biometric profiles have a close resemblance to the ground truth profiles which can lead to sensitive data exposure, like revealing the time of the day an individual wakes-up and goes to bed. Keywords Impersonator attack · Biometric profiles · Machine learning · Genetic algorithms
Introduction The use of wearable devices, sensing technologies and machine learning methods to monitor and predict user behavior has gained a lot of attention in recent years. It has been shown that these technologies have great potential to solve many relevant problems such as continuous mental health monitoring [22], elderly care assistance [44], cancer detection [26] and sports monitoring [4]. Many of those works use wearable sensors data, such as accelerometers, gyroscopes, temperature or heart rate, to train machine learning models. Being already embedded in many types of devices like smartphones, smart-watches and fitness bracelets, accelerometers are the most common wearable sensors. Actigraphy devices record human motion levels This article is part of the Topical Collection on Image & Signal Processing Enrique Garcia-Ceja
[email protected] 1
SINTEF Digital, Oslo, Norway
2
Barcelona Supercomputing Center, Barcelona, Spain
3
SimulaMet, Oslo, Norway
using accelerometers. Usually, these devices are worn on the wrist like a regular watch and are very popular in medical studies due to their non-invasive nature. Actigraphy devices can be used to monitor sleep patterns [25] or predict depression states [21], among other things. Recent works have suggested that such devices can be used to capture user behavior to build biometric profiles, i.e., be
Data Loading...
