A prediction model of an object variable with high accuracy is made by multiple regression analysis using a computer. (a) An initial sample set of samples for which the measured value of the object variable is known is prepared, (b) the set is subjected to multiple regression analysis to obtain calculated values of the object variable, (c) a sub-sample set of samples for each of which the difference between the measured value and the calculated value is a first value or less is subjected to multiple regression analysis to calculate a determination coefficient, (d) step (c) is repeated until the determination coefficient exceeds a second value while varying the first value, (e) two-class classification is carried out so that the sub-sample set after step (d) is used as a first sub-sample set and the other samples are grouped into a second sub-sample set to define a discriminant function, (f) the multiple regression equation used for analysis after step (d) and the discriminant function at step (e) are determined as a prediction model, and (g) the second sub-sample set is used as the initial sample set, and prediction models are made by repeating step (b) to step (f).
展开▼
