The dmcen.m function allows to compute the Diagonal Modified Confusion Entropy (DMCEN), which assess the performance of class-models jointly computed for k classes.
DMCEN is a versatile index regarding sensitivity (capability of each class-model to contain its own objects) and specificity (capability of the each class-model to reject foreign objects). DMCEN develops the idea that a classification model introduces an order in the objects of a dataset, capable of being measured by the decrease in entropy that it entails.
A detailed description of the algorithm and its properties for evaluating a k-class-model with respect to other indexes can be seen at:
O. Valencia M.C. Ortiz, M.S. Sánchez, L.A. Sarabia, A modified entropy-based performance criterion for class-modelling with multiple classes. Chemometrics and Intelligent Laboratory Systems 217 (2021) 104423. https://doi.org/10.1016/j.chemolab.2021.104423
- S: sensitivity/specificity matrix. The diagonal of S is made up of the sensitivity of each class.
- w: value between 0 and 1. It’s the weight for the convex combination between individual in-diagonal and the off-diagonal values of modified confusion entropy.
- dmcen: value of the k-class-model
- dmcenid: vector with the k individual DMCEN values for each class
The function code indicates the relationship of some sentences to the equations defined in the above-mentioned paper.
M.S. Sánchez, O. Valencia, S. Ruiz, M.C. Ortiz, L.A. Sarabia “DMCEN a MATLAB function to evaluate the entropy improvement provided by a multivariate k-class-model”