β Input |
ββA training set ; Number of input features contained in each feature |
ββ subset; A base learner ; Number of iterations ; A new data point to be classified. |
β Training Phase |
ββFor |
βββ Calculate the rotation matrix for the th classifier |
β β (1) Randomly split the feature set into subsets . |
β β (2) For |
ββ β β(a) Select the columns of that correspond to the attributes in to compose a |
βββββ submatrix . |
β β β β(b) Draw a bootstrap sample (with sample size smaller than that of , |
βββββgenerally taken to be 75%) from . |
ββ β β(c) Apply PCA to to obtain a matrix whose th column consists of the |
βββββ coefficients of the th principal component. |
β β β(3) EndFor |
β β β(4) Arrange the matrices into a block diagonal matrix . |
β β β(5) Construct the rotation matrix by rearranging the rows of so that |
βββββthey correspond to the original features in . |
βββ Provide as the input of to build a classifier . |
ββEndFor |
β Output |
βββ The class label for predicted by the ensemble classifier as |
βββββββββββββ. |