Title: Supervised classification using a distance-depth function
Authors: Itziar Irigoien - University Basque Country (Spain) [presenting]
Concepcion Arenas - University of Barcelona (Spain)
Francesc Mestres - University of Barcelona (Spain)
Abstract: Supervised classification is used by researchers in a wide variety of fields as in taxonomic classification; in morphometric analysis for species identification; in ecological problems addressed to test the presence or absence of a particular species; in marine ecology to evaluate the similarity of distinct populations and to classify units of unknown origin to known populations; in genetic studies in order to summarize the genetic differentiation between groups or in the biomedical context, predicting the diagnostic category of a sample on the basis of its gene expression profile and some clinical features. A novel classifier rule is introduced based on an improvement of the distance-based discriminant (DB-discriminant), taking into account a depth function. This new model combines the DB-rule and the maximal depth classifier, obtaining a classifier that is often more accurate than both methods separately. To demonstrate its effectiveness the new classifier was compared with the DB-rule and the $k-$nearest neighbor classification method, using high-dimensional class-imbalanced cancer data sets, and evaluating the leave-one-out error rate, the generalized correlation coefficient, the sensitivity, the specificity and the positive predicted value for each class. The results show the good performance of the new classifier.