Quality performance metrics in multivariate classification methods for qualitative analysis

Abstract

From an analytical point of view, this tutorial shows how to evaluate the quality of the multivariate classification methods used in qualitative analysis. This evaluation varies according to the type and number of the input and output classes involved. According to the variety of results that might be produced by a classification process, we define and explain the most suitable contingency charts for each type and depending on the analytical results of the validation step. These contingency charts provide a very useful alternative to the conventional contingency tables that are usually applied in this context.We define the concept of ‘classification scenario’ as a combination of the analytical and classification methods applied to each particular case. To conclude we make an exhaustive summary of the quality performance metrics. These metrics allow us to rank the different classification scenarios that appear in the usual analytical problems where different methods are compared.