DEFICIT VALUE PREDICTION DEVICE, DEFICIT VALUE PREDICTION METHOD AND DEFICIT VALUE PREDICTION PROGRAM

摘要

PROBLEM TO BE SOLVED: To select variables contributing to the improvement of predicting accuracy such that the quantity and quality of learning data are simultaneously satisfied for the purpose of improvement in the predicting accuracy of a deficit value in imperfect data.SOLUTION: In order to highly accurately predict the deficit value of imperfect data, a deficit value prediction device alternately repeats two processing of removing variables whose correlation with a deficit value to be predicted is low among variables from learning data to increase the quality of the learning data, and setting data in which the values of the residual variables are aligned as new learning data to increase the quantity of the learning data such that it is possible to narrow down the variables to those contributing to the improvement of predicting accuracy.