Abstract: A response surface model is created without a drawback of creating a linear model that does not approximate to samples belonging to a partitioned region. Provided are: an input unit 1 that inputs samples composed of factor values and a response value; a sample placement unit 21 that places the samples in spaces, the samples being input by way of the input unit 1; a linear modeling unit 230 that creates the linear model for each region, based on coordinate values of the samples belonging to the region; a region partition unit 22 that partitions the linear model, based on the samples belonging to the region; a partition region determination unit 221 that determines whether it is possible to partition the region, based on the samples belonging to the region to be partitioned; and a modeling unit 23 that creates a response surface model by placing each linear model in the spaces when the partition region determination unit 221 has determined that partition is impossible in all the regions.

Abstract: A response surface model is created without a drawback of creating a linear model that does not approximate to samples belonging to a partitioned region. Provided are: an input unit 1 that inputs samples composed of factor values and a response value; a sample placement unit 21 that places the samples in spaces, the samples being input by way of the input unit 1; a linear modeling unit 230 that creates the linear model for each region, based on coordinate values of the samples belonging to the region; a region partition unit 22 that partitions the linear model, based on the samples belonging to the region; a partition region determination unit 221 that determines whether it is possible to partition the region, based on the samples belonging to the region to be partitioned; and a modeling unit 23 that creates a response surface model by placing each linear model in the spaces when the partition region determination unit 221 has determined that partition is impossible in all the regions.