PURPOSE: To reduce learning data obtained by selecting a sample for increasing a recognition rate among all collected learning data by selecting a subgroup as learning sample data by a method using a distance function based upon multivariable and phase geometric analysis and an automatic classifying method.
CONSTITUTION: A neural network which uses sample data as learning data is determined and converged by learning, the recognition rate of test data among all sample data is observed first, and an optional number of groups (cluster) are generated according to distances between samples; and each cluster or specific clusters are learnt and converged as new learning data, and a subgroup of sample data is selected by using the recognition rate of the test data as an evaluation reference. In this case, the subgroup is selected by the method using multivariable and phase geometric analysis and the automatic classifying method. Therefore, the size of the learning data is reducible without lowering the recognition rate of the neural network.