The present invention relates to a lightweight random forest classification method by applying a soft target learning method, and more specifically, as a random forest classification method, (1) using a date set A, applying a conventional random forest learning method to a teacher Learning a random forest; (2) using the teacher random forest learned in step (1), extracting probability values of each class constituting the data set B for the student random forest; (3) using the data set B from which the probability values of each class were extracted in step (2), learning a student random forest; And (4) performing classification using the Student random forest learned in step (3). In addition, the present invention relates to a classifier using a lightweight target classification method by applying a soft target learning method, and more specifically, as a classifier using a random forest classification method, (1) using a date set A, existing A teacher random forest learning module for learning a teacher random forest by applying a random forest learning method; (2) a class probability value extraction module for extracting a probability value of each class constituting the data set B for a student random forest using the teacher random forest learned in the teacher random forest learning module; (3) a student random forest learning module that trains a student random forest using a data set B from which the probability values of each class are extracted from the class probability value extraction module; And (4) a classification module that performs classification using the Student random forest learned in the Student random forest learning module. According to the classification method using the soft target learning method proposed in the present invention and the lightweight random forest classification method and the classifier using the same, the teacher random forest is trained using the existing random forest learning method, and the teacher random forest thus trained After extracting the probability values of each class constituting the data set for the student random forest, using the data set from which the probability values of each class are extracted to train the student random forest, the random forest performance is maintained while random By reducing the number of trees in the forest, processing time and amount of memory can be significantly reduced.
展开▼