Behavioural biometrics: a survey and classification

Roman V. Yampolskiy, Venu GovindarajuInt. J. Biom 2008 · DOI

TL;DR

This paper surveys and classifies the burgeoning field of behavioral biometrics, which uses unique human actions and patterns for identification. It categorizes existing methods, analyzes their features and performance, and discusses the critical privacy implications of this technology.

What Problem Does This Paper Address?

As digital systems become more pervasive, robust and user-friendly security measures are paramount. Traditional biometrics (fingerprints, iris scans) are well-established but often require dedicated hardware and can be intrusive. Behavioral biometrics offer a compelling alternative by leveraging inherent human actions (like typing patterns, gait, or voice) that can be captured non-intrusively, often with existing hardware. However, the field is fragmented, with various approaches and classifications. This paper aims to bring order to this landscape by providing a comprehensive survey and a structured classification of behavioral biometrics, highlighting their strengths, weaknesses, and future potential.

Methods

The authors propose a five-category classification for behavioral biometrics based on the type of information collected:

  1. Authorship-based: Analyzing text or drawings for stylistic peculiarities.
  2. Human-Computer Interaction (HCI)-based: Quantifying user interaction with computers. This is further subdivided into:
  3. Input device interaction: Analyzing actions with keyboards, mice, etc.
  4. Advanced human behavior: Measuring strategy, knowledge, or skill during interaction.
  5. Indirect HCI-based: Monitoring low-level system actions (e.g., system calls, audit logs) that indirectly reflect user behavior.
  6. Motor-skill based: Measuring innate, unique, and stable muscle actions.
  7. Purely behavioral: Quantifying human behavior not directly tied to specific motor actions, focusing on strategies and knowledge.

The paper then details numerous specific behavioral biometrics within these categories, discussing the features extracted (e.g., keystroke timing, mouse movement patterns, gait parameters, voice spectrum) and the underlying techniques used for analysis and verification. A generalized algorithm for behavioral biometrics is also outlined, involving breaking down behavior into atomic actions, creating feature vectors, and using similarity measures for verification.

Datasets

The paper does not introduce new datasets. Instead, it surveys existing research and refers to datasets used in those studies. Specific datasets are not detailed, but the references point to various sources used for training and evaluating different behavioral biometric methods.

Key Insights and Ideas

  • Ubiquitous Data Capture: Behavioral biometrics can often be collected passively and non-intrusively, even without the user's explicit knowledge, making them highly acceptable.
  • Cost-Effectiveness: Many behavioral biometrics can be implemented using existing hardware, reducing the cost barrier compared to specialized biometric systems.
  • Verification vs. Identification: While most behavioral biometrics are currently suitable for user verification (confirming identity), only a few (like signature/handwriting and speech) show promise for large-scale user identification.
  • Privacy Concerns are Paramount: A significant insight is the inherent risk of behavioral profiles revealing sensitive personal information (e.g., health, habits, preferences) beyond just identity. This necessitates robust encryption and ethical considerations.
  • The "Soft Biometrics" Complement: Traits like height, gender, or even IQ scores (soft biometrics) are not sufficient for identification but can be combined with other biometrics to enhance accuracy or reject false claims.
  • Concept Drift is a Challenge: Human behavior evolves over time as users learn and adapt, posing a challenge for the long-term permanence of behavioral biometric models.

Key Results

The paper presents a comprehensive overview of various behavioral biometrics, categorizing them and detailing their features. Table 4 and Table 5 provide a summary of recognition, verification, and error rates for many of these biometrics, citing specific publications. For example:

  • Credit card use shows very high detection rates (99.995%) with a low false rejection rate (20%).
  • GUI interaction demonstrates high detection rates (96.15%) with a low false positive rate (3.85%).
  • Keystroke dynamics achieves high accuracy (95.7%) with low error rates (2.3%).

Limitations and Future Work

  • Universality and Uniqueness: The paper notes that the universality of behavioral biometrics is low in a general population but 100% within a specific domain. Uniqueness is also relatively low, often only sufficient for verification in small user groups.
  • Permanence: Behavioral biometrics can exhibit a low degree of permanence due to concept drift (users learning and changing their behavior). However, research is ongoing to develop adaptive systems that can adjust to these changes.
  • Circumvention: While difficult, it is possible to circumvent behavioral biometric systems with intimate knowledge of another person's behavior. This highlights the critical need for secure storage and encryption of behavioral profiles.
  • Future Work: The authors suggest that increasing overall accuracy through multimodal behavioral biometrics (combining multiple behaviors) and exploring new applications in areas like personalized education, mobile commerce, and risk analysis are promising directions. Addressing privacy concerns remains a crucial area for future research and development.

Summary updated

BibTeX
@article{yampolskiy2008behavioural,
  author = {Roman V. Yampolskiy and Venu Govindaraju},
  title = {{Behavioural biometrics: a survey and classification}},
  journal = {Int. J. Biom.},
  year = {2008},
  doi = {10.1504/ijbm.2008.018665},
}
Figure