CONSTITUTION METHOD FOR MAJORITY DECISION PREDICTION MACHINE, MAJORITY DECISION PREDICTION MACHINE AND RECORDING MEDIUM THEREFOR

摘要

PROBLEM TO BE SOLVED: To reduce an error probability to unknown data in the case of linearly connecting plural prediction machines (base classifiers) by a weighted majority decision method and constituting a majority decision prediction machine. SOLUTION: A weight p(h) to the base classifier is decided at random (step 101), the minimum value of an error rate estimated based on the p (h) decided in such a manner and the output h (X) of the respective base classifiers is obtained (step 102) and whether or not the operations of the step 101 and the step 102 are repeated the sufficient number of times is judged (step 103). In the case that they are not repeated the sufficient number of times, the step 101 is returned and the processing is repeated. In the case that they are repeated the sufficient number of times, the smallest one of the error probabilities estimated in the repetitions is defined as final output (step 104).