Up to know, several statistical classifiers are implemented. MDBC implemented the mahalanobis distances based classifier, GDBC is a general discriminant-based classifier supporting several variants of quadratic classifiers.
LDBC implemented a two-class linear discriminant analysis (LDA), LDBC2, LDBC3 and LDBC4 are multi-class LDA with slightly different covariance matrix.
Due to user demand, a separate training and testing functions was implemented. The training function TRAIN_SC is simple and the same for all these statistical classifiers. Basically, only the mean and the covariance matrices for each class have to be calculated. Within BIOSIG, an efficient encoding using the so-called "extended covariance matrix" is chosen.
The testing function is called TEST_SC, the third argument determines which classifier is used. If the target_classlabels are used as the forth input argument, also the classification accuracy, the confusion matrix and Cohen's kappa coefficient are calculated.
Here is a short demo with some random data. >> CC=train_sc(randn(1000,4),mod(1:1000,4)') CC = Labels: [4x1 double] datatype: 'classifier:statistical' MD: [4x5x5 double] >> R=test_sc(CC,randn(1000,4),'LD2',mod(1:1000,4)') Warning LDBC: 1-column added to data R = output: [1000x4 double] classlabel: [1000x1 double] kappa: -0.0267 H: [4x4 double] ACC: 0.2300 For more details, see HELP TRAIN_SC and HELP TEST_SC.The functions are available in the CVS-repository and will be included in the next release of BIOSIG