Analysis of errors of handwritten digits made by a multitude of classifiers

In this paper we describe an in-depth study on some data misclassified by a collection of classifiers produced by different authors. First of all, we divide the errors into three categories based on their quality and analyze their distributions according to category. Common errors made by three or more classifiers out of five have been identified and analyzed to deduce the reasons of misclassification. Finally, based on systematic analyses, two possible solutions to reduce errors and improve system reliability are proposed: (a) a verification module, and (b) combination of complementary multiple classifiers.