An effective reject rule for reliability improvement in bank note neuro-classifiers

In this paper the reliability of bank note neuro-classifiers is investigated and a reject rule is proposed on the basis of probability density function of the input data. The reliability of classification is evaluated through two parameters, which are associated with the winning class probability and the second maximal probability. Then a threshold value is considered to reject the unreliable classifications. As for modeling the non-linear correlation among the data variables and extracting the features, a local principal components analysis (PCA) is applied. The method is tested with a learning vector quantization (LVQ) classifier using 3,600 data samples of various bills of US dollar. The results show that by taking a suitable reject threshold value and also a proper number of regions for the local PCA, the reliability of the system can be improved significantly.