PROBLEM TO BE SOLVED: To raise the accuracy of extracting an other time-series that relates to some time-series.SOLUTION: A correlation factor extraction method of the present invention includes the steps of: classifying into a first group at least some of prediction models, among a plurality of prediction models generated from one first time-series data and a plurality of second time-series data, the estimated values of prediction error of which are smaller than the estimated values of prediction error of a reference prediction model generated from the first time-series data; classifying into a second group at least some of the remaining prediction models not classified into the first group; calculating the probability of the prediction error of the prediction models classified into the second group becoming smaller than the prediction error of the reference prediction model on the basis of a condition that classification into the first group is correct, using the prediction value of the prediction model classified into the second group; and determining whether or not the prediction model classified into the second group should be reclassified into the first group on the basis of whether or not the calculated probability is larger than a prescribed threshold.SELECTED DRAWING: Figure 7
展开▼