DCSC

Dynamic Classifier Selection Combiner

    V = DCSC(A,W,K,TYPE)
    V = A*(W*DCSC([],K,TYPE))
    D = B*V

Input
A	Dataset used for training base classifiers as well as combiner
B	Dataset used for testing (executing) the combiner
W	Set of trained or untrained base classifiers
K	Integer, number of neighbors
TYPE	'soft' (default) or 'crisp'

Output
V	Trained Dynamic Classifier Selection
D	Dataset with prob. products (over base classifiers) per class

Description

This dynamic classifier selection takes for every object to be  classified the K nearest neighbors of an evaluation set (training set) A and determines which classifier performs best over this set of objects.  If the base classifiers (STACKED or PARALLEL) are untrained, A is used to  train them as well.

The selection of the best classifier can be made in a soft or in a crisp  way. If TYPE is 'soft' (default) classifier confidences are averaged (see  CLASSC), otherwise the best classifier is selected by voting.

Reference(s)

G. Giacinto and F. Roli, Methods for Dynamic Classifier Selection 10th Int. Conf. on Image Anal. and Proc., Venice, Italy (1999), 659-664.

See also

datasets, mappings, stacked, naivebc, classc, testd, labeld,