Predictive learning by ensemble learning of heterogeneous machine learning

摘要

Problem to be solved: to provide a bank prediction system for estimating the probability of a rare event such as a default rate of a bank loan with a small data set.The predictor isA model learning section 102 for learning a new bank prediction model from training data of a new bankBased on the first bank model of the first existing bankA first predictor 402A outputting a first prediction resultBased on the second bank model of the second bankA second predictor 402b outputting a second prediction resultBased on new bank validation data and a new bank prediction modelA third predictor 402c outputting a third prediction resultA gate ensemble learning unit 531 that learns the gate ensemble weights such that the ensemble results of the first to third prediction results approach the desired variable of validation dataA storage device for storing learned gate ensemble weight values.Diagram