A Review of Several Keystroke Dynamics Methods

TL;DR

This paper compares various keystroke dynamics authentication methods using statistical models and timing features, evaluating their effectiveness in distinguishing genuine users from impostors across multiple datasets.

Contribution

It provides a comparative analysis of GMM, Mahalanobis distance, and Gunetti Picardi's metrics for keystroke authentication, highlighting their relative performance.

Findings

GMM achieved the lowest EER among models.

Mahalanobis distance showed high accuracy with certain features.

All models demonstrated varying effectiveness depending on dataset and features.

Abstract

Keystroke dynamics is a behavioral biometric that captures an individual's typing patterns for authentication and security applications. This paper presents a comparative analysis of keystroke authentication models using Gaussian Mixture Models (GMM), Mahalanobis Distance-based Classification, and Gunetti Picardi's Distance Metrics. These models leverage keystroke timing features such as hold time (H), up-down time (UD), and down-down time (DD) extracted from datasets including Aalto, Buffalo and Nanglae-Bhattarakosol. Each model is trained and validated using structured methodologies, with performance evaluated through False Acceptance Rate (FAR), False Rejection Rate (FRR), and Equal Error Rate (EER). The results, visualized through Receiver Operating Characteristic (ROC) curves, highlight the relative strengths and weaknesses of each approach in distinguishing genuine users from impostors.