Peer-reviewed research
Comparing Anomaly-Detection Algorithms for Keystroke Dynamics
Kevin S. Killourhy and Roy A. Maxion, Carnegie Mellon University, Dependable Systems Laboratory, 2009
Read the originalRead on October 16, 2026
- What it is
- Carnegie Mellon researchers collected typing timings from 51 people, each typing the same password 400 times, and compared 14 detectors of impostors on that one dataset. They released the data and a repeatable test procedure, which later studies use as a shared benchmark.
- How much weight it carries
- Peer-reviewed and reused by many later papers. The task was one fixed password typed in a lab, so the error rates describe that setup, not free typing in everyday life.
Facts taken from this source
- The Carnegie Mellon keystroke benchmark collected timings from 51 people, each typing a password 400 times.
“We collected data from 51 subjects typing 400 passwords each” (Abstract)
- The three top-performing of 14 keystroke-dynamics detectors reached equal error rates between 9.6 and 10.2 percent on that benchmark.
“The three top-performing detectors achieve equal-error rates between 9.6% and 10.2%.” (Abstract)
- The Carnegie Mellon keystroke study released its data and evaluation method, so other researchers can compare detectors on the same dataset.
“along with the shared data and evaluation methodology” (Abstract)