In this study, the application of classifiers to problems in fingerprint prediction is investigated. Six supervised learning and two statistical fingerprint classification methods (classifiers) are considered: linear discriminant analysis (LDA); logistic discrimination (LgD), k-nearest neighbour (k-NN), artificial neural network (ANN), association rules (AR) decision tree (DT), naive Bayes classifier (NBC) and the support vector machine (SVM). The performance of several multiple classifier systems are also demonstrated and evaluated in terms of their ability to correctly predicting or classifying a fingerprint using the National Institute of Standards and Technology (NIST) biometric image database. Examining the performance of the base classifiers showed DT, SVM and ANN to have the highest accuracy while LDA has the lowest accuracy rate. The results further show all the multi stage systems to significantly outperform the baseline classifiers. Accordingly, good performance is consistently derived from boosting.
Reference:
Molale, P, Twala, B and Seeletse, S. 2011. Fingerprint prediction using classifier ensembles. 53rd Annual Conference of the South African Statistical Association for 2011 (SASA 2011), CSIR, CSIR Convention Centre, Pretoria, South Africa, 1-3 November 2011
Molale, P., Twala, B., & Seeletse, S. (2011). Fingerprint prediction using classifier ensembles. South African Statistical Association for 2011. http://hdl.handle.net/10204/5379
Molale, P, B Twala, and S Seeletse. "Fingerprint prediction using classifier ensembles." (2011): http://hdl.handle.net/10204/5379
Molale P, Twala B, Seeletse S, Fingerprint prediction using classifier ensembles; South African Statistical Association for 2011; 2011. http://hdl.handle.net/10204/5379 .
53rd Annual Conference of the South African Statistical Association for 2011 (SASA 2011), CSIR, CSIR Convention Centre, Pretoria, South Africa, 1-3 November 2011