An Efficient User Verification System via Mouse

• Background Knowledge and Insights
• Goals
• Data
  • Data Source
  • Data Processing
• Feature(Metrics)
• Mouse Movement Characterization
  • Dependence on different platforms
  • Distance Between Distributions
  • Number of Mouse Clicks in a Real Session
• Classifier
• Limitation
• References

Background Knowledge and Insights

• Re-verification system
  • Accuracy
  • Quick response
  • Difficult to forge normal biometric behaviors
• Frequent User verification should be
  • Passive
  • Transparent to users
• Shortcomings of some behavioral biometrics approaches
  • Fingerprints, retinal scan
    • Specialized hardware
      • Expensive
      • Unavailable
  • Keystroke
    • Record sensitive user information
      • Username
      • Password
    • Complex structure (shape, size, layout)
• Angle-based metrics
  • Reduces verification time
  • High accuracy
  • Independent of the operating environment

Goals

• A Novel measurement strategy, angle-based metrics
• An experiment involving sessions from over 1,000 unique users

Data

Data Source

• Two data sets
  • Controllable environment
    • Controllable set
    • 30 users
      • Different backgrounds
  • Online forum
    • Field set
    • 1000 real field users
    • Recorded by JaveScript code
• Raw data
  • $\langle \textrm{ACTION-TYPE}, t, x, y \rangle$
    • $\textrm{ACTION-TYPE}$, {mouse-move, mouse-click}
    • $t$, timestamp of the mouse action, collected in milliseconds
    • $x$, $y$, coordinate

Data Processing

• Identify every point-and-click action
  • Continuous mouse movement followed by click
  • $\langle \textrm{mouse-move}, t_i, x_i, y_i\rangle_{c, j}$
    • $i$, $i^{th}$ point-and-click action
    • $c$, user
    • $j$, $j^{th}$ mouse move record
    • $t_i$, timestamp

Feature(Metrics)

• Direction
  • For consecutive recorded points $A$, $B$: $\vec{AB}$
• Angle of Curvature
  • For any three consecutive points $A$, $B$, $C$: $\angle{ABC}$, angle between $\vec{AB}$ and $\vec{BC}$
• Curvature Distance
  • For any three consecutive points $A$, $B$, $C$: ratio between $\vert\vec{AC}\vert$ to length of perpendicular distance from $B$ to $\vec{AC}$
• Speed
  • the total distance traveled for that action $/$ the total time taken to complete the action
• Pause-and-Click
  • Time between the end of the movement and the click event

Mouse Movement Characterization

Dependence on different platforms

• OS
• Screen
  • Size
  • Resolution
• Mouse
  • Mouse pointer sensitivity
  • Brand of mouse
• Desk space available near mousepad
• Poor feature choices
  • Speed
  • Acceleration
  • Pause-and-click
    • Dependent on the reading content
• Uniqueness of angle-based metrics across users

Distance Between Distributions

• Angle-based features are continuous variables
• Divided into discrete intervals, $bins$
• Calculate PDF for each distribution
• $\mathrm{PDF}_p = \{p_1, p_2, ..., p_n\}$
  • $\mathrm{PDF}_p$ for distribution $p$
  • $p_i$ represents the probability of falling into the $bin_i$
• Distance between $\mathrm{PDF}_p$, $\mathrm{PDF}_q$
  • $D(p,q)=\sum_i^n \vert p_i-q_i\vert$

Number of Mouse Clicks in a Real Session

• Average number of mouse clicks per session being about 15
• User must be identified in fewer than 15 clicks

Classifier

• 2-class SVM
  • RBF kernel
• Decision
  • Threshold
  • Majority vote

Limitation

• Partial Movements
  • Continuous mouse movements without ending in a click
    • Aimless
      • Move its mouse just to stop the screen saver when watching a video
    • Intentionally performed
      • Aid reading
      • Moving the mouse to a link, but then decide not to click on it
  • Compare to point-and-clicks
    • More noisy
    • Much more frequent
      • 0.53 mouse clicks per minute
      • 6.58 partial movements per minute
  • Reduce verification time, at the cost of accuracy degradation

• Scalability problem
  • Common problem for almost all biometrics approaches
• More suitable to work together with other authentication methods

References

• An Efficient User Verification System via Mouse Movements, 2011
• CS 259D Lecture 7