Related Experiment Video
Updated: Jul 7, 2026
05:42
An Assessment Method and Toolkit to Evaluate Keyboard Design on Smartphones
Published on: October 5, 2020
Verification of computer users using keystroke dynamics
1Dept. of Comput. Sci., Monmouth Univ., West Long Branch, NJ.
Summary
Keystroke dynamics, including key hold times, enhance computer user verification. Combining interkey and hold times with specific neural networks achieved 100% accuracy in identifying users by their typing patterns.
Area of Science:
- Computer Science
- Biometrics
- Pattern Recognition
Background:
- Previous research utilized interkey times for computer user identification.
- Current study explores key hold times and combined temporal features for enhanced security.
Purpose of the Study:
- To investigate the efficacy of key hold times versus interkey times for user authentication.
- To develop and compare neural network and pattern recognition techniques for keystroke dynamics analysis.
- To achieve high-accuracy computer user verification through advanced biometric methods.
Main Methods:
- Utilized key hold times as primary features for classification.
- Compared hold time performance against previous interkey time-based methods.
- Integrated both interkey and hold times for a comprehensive identification approach.
- Applied various neural network paradigms (e.g., Fuzzy ARTMAP, RBFN, LVQ) and classical algorithms.
Main Results:
- Key hold times demonstrated superior performance compared to interkey times alone.
- The combined approach using both interkey and hold times yielded the highest identification accuracy.
- Achieved 100% identification accuracy with Fuzzy ARTMAP, RBFN, and LVQ using combined temporal features.
- Other algorithms like backpropagation and Bayes' rule showed moderate performance.
Conclusions:
- Key hold times are more effective than interkey times for user authentication.
- Combining interkey and hold times significantly improves the accuracy of keystroke dynamics-based identification.
- Advanced neural network models, particularly Fuzzy ARTMAP, RBFN, and LVQ, are highly effective for biometric user verification using typing patterns.
