EVALUATING PATTERN RECOGNITION PROBLEM
Władysław Homenda, Witold Pedrycz · Pattern Recognition · 2018
This chapter defines various factors and measures in order to provide ways to reliably evaluate the quality of pattern recognition without and with rejection mechanisms. It illustrates several experimental studies in which the authors present how to evaluate the quality of native pattern recognition with foreign pattern rejection. For better understanding of how quality of classification with rejection should be measured, parameters and quality measures are adopted that are used in signal detection theory and in statistics. Classification with rejection is aimed at the maximization of all mentioned measures. Those measures complete discrimination between native and foreign patterns and classification of native patterns into respective classes. The chapter presents a confusion matrix, which serves as a way to visually present the results. It presents an evaluation for two pattern processing schemes: pure classification (without rejection) and classification with rejection applied to native patterns only.