BAGGINGC

Bootstrapping and aggregation of classifiers

     W = BAGGINGC (A,CLASSF,N,ACLASSF,T)

Input
A	Training dataset.
CLASSF	The base classifier (default: NMC)
N	Number of base classifiers to train (default: 100)
ACLASSF	Aggregating classifier (default: MAXC).
T	Tuning set on which ACLASSF is trained (default: [], meaning use A)

Output
W	A combined classifier (if ACLASSF given) or a stacked  classifier (if ACLASSF []).

Description

Computation of a stabilised version of a classifier by bootstrapping and  aggregation ('bagging'). In total N bootstrapped versions of the dataset A are generated and used for training of the untrained classifier CLASSF.  Aggregation is done using the combining classifier specified in ACLASSF.  If ACLASSF is a trainable classifier it is trained by the tuning dataset  T, if given; else A is used for training. The default aggregating classifier  ACLASSF is MAXC. Default base classifier CLASSF is NMC.

In case the aggregating classifier ACLASSF is WVOTEC the weights for the  voting are derived from the apparent errors based on the bootstrapped  versions of the training set A.

In multi-class problems another way of combining might be of interest
W = A*(BAGGINGC*QDC([],[],1e-6)).

Reference(s)

L.Beiman, Bagging Predictors, Machine Learning, 1996.

See also

datasets, mappings, nmc, maxc,