Approximate Confidence Interval Estimation for Beta-binomial Proportions of Biometric Identification Devices
Travis Jason Atkinson
Dr. Michael Schuckers, Faculty Mentor
In this talk we consider methods for making a confidence interval for error rates of a biometric identification device. These devices, such as fingerprint scanners or iris scanners, are becoming increasingly prevalent. We present and evaluate four new approaches to confidence interval estimation of the matching error rates of biometric identification devices. These matching errors often follow a Beta-binomial distribution. Therefore, we propose extensions to the methodology of Agresti and Coull for the Beta-binomial distribution. We compare these methods using a Monte Carlo simulation and make recommendations for future work.