Smartwatch User Authentication by Sensing Tapping Rhythms and Using One-Class DBSCAN

Hanqi Zhang, Xi Xiao, Shiguang Ni, Changsheng Dou, Wei Zhou, Shutao XiaSensors 2021 · DOI · OpenReview

TL;DR

This paper introduces a novel smartwatch authentication method using user tapping rhythms and a new One-Class DBSCAN algorithm. It achieves a significantly lower Equal Error Rate (EER) of 0.92% compared to existing methods, enhancing security and privacy for wearable devices.

What Problem Does This Paper Address?

Smartwatches, increasingly prevalent in our lives, collect sensitive personal data. Protecting this data from unauthorized access is crucial, but existing authentication methods for wearables often fall short. Traditional methods like passwords and patterns are cumbersome on small smartwatch screens, while biometric approaches can require specialized sensors or high energy consumption. Behavioral biometrics, while promising, can also be power-intensive or require specific user actions. This paper addresses the need for a lightweight, secure, and user-friendly authentication system for smartwatches that can be implemented on the device itself (edge computing) to minimize privacy risks.

Methods

The core of this paper lies in two main contributions:

  • Feature Extraction: The authors propose a method to extract features from tapping rhythms. Each tap sequence is represented as a vector of durations and intervals between taps. To make these features robust to variations in tapping speed or user urgency, they are normalized by the total duration of the rhythm. This creates a more stable representation for classification.
  • One-Class DBSCAN: To address the challenge of one-class classification (where only legitimate user data is available for training), the authors adapt the DBSCAN clustering algorithm. Instead of identifying multiple clusters, their "One-Class DBSCAN" aims to identify a single cluster representing the legitimate user's tapping rhythm. It first identifies "core objects" within the training data and then uses these to define the boundaries of the legitimate class. New tapping rhythms are authenticated if they fall within the defined cluster. This approach is designed to be robust to outliers and does not require negative (imposter) data for training.

Additionally, the paper proposes a method to detect the security level of a tapping rhythm password by analyzing the standard deviation of the tapping data, prompting users to create more complex patterns.

Datasets

The authors collected their own dataset due to the lack of public tapping rhythm datasets for smartwatches.

  • Dataset: Data was collected from a Moto 360 smartwatch using a custom Android application.
  • Collection: Each user entered their chosen tapping rhythm 10 times.
  • Size: A total of 6110 data samples were collected from over 600 users.
  • Focus: For experiments, datasets with tapping rhythm lengths of 5 to 8 were primarily used, as shorter or longer sequences were deemed potentially less reliable.

Key Insights and Ideas

  • Tapping Rhythms as a Viable Authentication Factor: The paper strongly advocates for tapping rhythms as a secure and practical authentication method for smartwatches, highlighting its advantages in terms of speed, power consumption, and ease of use compared to other methods.
  • One-Class Classification is Key: The inherent nature of user authentication (knowing what a legitimate user does rather than what all possible impostors do) makes one-class classification a natural fit, and the proposed One-Class DBSCAN effectively addresses this.
  • Feature Normalization Matters: Transforming raw tap durations and intervals into ratios relative to the total rhythm duration significantly improves the robustness and accuracy of the authentication system.
  • DBSCAN's Adaptability: The successful adaptation of DBSCAN, a density-based clustering algorithm, for one-class classification demonstrates its flexibility beyond traditional multi-cluster identification.
  • Security Beyond Just Recognition: The proposed method for assessing tapping rhythm security adds a crucial layer, guiding users towards stronger, less predictable patterns.

Key Results

  • Lowest EER: The proposed One-Class DBSCAN method achieved an Equal Error Rate (EER) of 0.92%, which is significantly lower than the 4.04% achieved by reproducing the method of Ben Hutchins et al. [27] on their dataset.
  • Feature Importance: Experiments showed that the authors' proposed features, when used with the One-Class DBSCAN algorithm, outperformed using the features from Ben Hutchins et al. [27] with their own algorithm (EER of 0.92% vs. 2.51%).
  • Algorithm Performance: When the authors' features were used with the Vector Comparison algorithm from Ben Hutchins et al. [27], the EER was 1.06%, still slightly higher than their proposed method.
  • Comparison with Other Algorithms: One-Class DBSCAN significantly outperformed Mean Shift (EER 53.4%) and Isolation Forest (EER 30.6%) for this task. Supervised learning methods like Decision Tree and Logistic Regression also showed poorer performance in terms of FAR/FRR balance.
  • Running Time: The average training time was 52.45 ms, and the average authentication time was a mere 10.58 ms, demonstrating the method's efficiency for edge computing on smartwatches.

Limitations and Future Work

  • Limitations:
    • The study acknowledges that tapping habits can change over time, potentially affecting authentication accuracy.
  • While the dataset was collected from real users, it's a single dataset, and generalization to diverse user populations and devices might require further validation.
  • The paper notes that supervised learning models struggled with the one-class nature of the data, highlighting the advantage of their unsupervised approach.
  • Future Work:
    • Multiple Tapping Rhythms: The authors aim to extend the method to recognize multiple different tapping rhythms for a single user, allowing for more flexible authentication.
  • Incremental Learning: They propose applying incremental learning to update the authentication model with users' current tapping rhythms, adapting to evolving habits.
  • Broader Applications: The authors believe their flexible and feasible approach can be applied to other scenarios with limited data and aim to explore these applications.
  • Periodic Reminders: Suggestion to periodically remind users to reset their tapping rhythm passwords to adapt to new habits.

Summary updated

Abstract

As important sensors in smart sensing systems, smartwatches are becoming more and more popular. Authentication can help protect the security and privacy of users. In addition to the classic authentication methods, behavioral factors can be used as robust measures for this purpose. This study proposes a lightweight authentication method for smartwatches based on edge computing, which identifies users by their tapping rhythms. Based on the DBSCAN clustering algorithm, a new classification method called One-Class DBSCAN is presented. It first seeks core objects and then leverages them to perform user authentication. We conducted extensive experiments on 6110 real data samples collected from more than 600 users. The results show that our method achieved the lowest Equal Error Rate (EER) of only 0.92%, which was lower than those of other state-of-the-art methods. In addition, a statistical method for detecting the security level of a tapping rhythm is proposed. It can prevent users from setting a simple tapping rhythm password, and thus improve the security of smartwatches.
BibTeX
@article{zhang2021smartwatch,
  author = {Hanqi Zhang and Xi Xiao and Shiguang Ni and Changsheng Dou and Wei Zhou and Shutao Xia},
  title = {{Smartwatch User Authentication by Sensing Tapping Rhythms and Using One-Class DBSCAN}},
  journal = {Sensors},
  year = {2021},
  doi = {10.3390/s21072456},
}
Figure