KeystrokeDynamics

A customisable authentication system based on keystroke biometrics, with a visualisation dashboard.

• KeystrokeDynamics
  • Screenshots
  • Introduction
  • How does it work?
    • Keystroke features
  • The detectors
    • Manhattan (Individual)
    • Manhattan Filtered (Individual)
    • Manhattan (Population)
    • Manhattan Filtered (Population)
    • Mahalanobis
  • Tech Stack

---

Screenshots

---

Introduction

Keystroke Dynamics or Biometrics is the manner and rhythm in which an individual types characters on a keyboard. These details are used to develop a unique biometric template of the user's typing pattern for future authentication.

How does it work?

• A basic keystroke logger implemented in keystrokeLogger.js and used in the forms
• Keystroke timings are recorded and sent to the server along with threshold details
• Server processes input and checks it against the data in the DB

Keystroke features

• Hold time (keydown-keyup)
• Flight time (keyup-keydown)
• DD time (keydown-keydown)

---

The detectors

Manhattan (Individual)

• Uses Manhattan distance between the mean and the input of each feature
• Allowed range is mean ± multiplier * SD
• If a certain percentage of input features fall in this range, the attempt is accepted
• Most effective when dealing with a small amount of data in the DB (<5 attempts) at default thresholds
• User controls:
  • SD Multiplier
  • Acceptance percentage

Manhattan Filtered (Individual)

• Same as the Manhattan detector, but with the outliers filtered out
• Allowed data is in the range mean ± 2.5 * SD
• Allows for a stricter dataset to compare against. Effective with >5 entries in the database at default thresholds
• User controls:
  • SD Multiplier
  • Acceptance percentage

Manhattan (Population)

• Uses Manhattan distance between the mean vector and the input vector
• Responds to minor deviations in input patterns
• Default thresholds are very strict
• Works well a decent amount of data (<10 attempts)
• User Controls
  • Distance threshold

Manhattan Filtered (Population)

• Same as the Manhattan detector, but with the outliers filtered out
• Responds to minor deviations in input patterns
• Default thresholds are very strict
• Works well a decent amount of data (<10 attempts)
• User Controls
  • Distance threshold

Mahalanobis

• Uses the Mahalanobis distance between the mean and input vectors
• Quite sensitive to deviations in input patterns
• Default thresholds are not too strict
• Can be used in combination with the Manhattan detectors for a precise decision
• User Controls
  • Distance threshold

---

Tech Stack

• Node.JS
• MongoDB
• HTML
• CSS (halfmoon)
• JS