The Power of Ensemble Models in Fingerprint Classification: A case study

Abstract

The usage of fingerprints as biometrics has been practiced for more than 100 years, with the popularization of sensors and fingerprint capturing methodologies, the usage of this method for authentication and recognition has grown in the past years. However, the usage for recognition in large databases with a huge number of entries is computationally costly, hence the classification of fingerprints aims to attenuate this cost by increasing optimization. This paper presents a performance comparison between two ensemble of classifiers and a decision tree classifier, applied to the database from a known benchmark, the NIST sd-14 database, for the classification of fingerprints. The comparison performed using the stratified cross-validation process to set confidence interval for the evaluation of performance measured by the success rate, using a Random Forest, XGBoost and Decision Tree as classifiers. The one-tailed paired t-test showed that Random Forest and XGBoost don’t have statistical difference with significance of 95%, however, their performance is superior to the simple classifier Decision Tree.