PURPOSE: To reduce an error between a forecasting value and an experience value by identifying which past data pattern corresponds to the pattern of the input data to be forecasted and using a forecasting device appropriate for the pattern of input data to be forecasted.

CONSTITUTION: A pattern identifying part 2 judges which past actual data pattern corresponds to the pattern of input data to be forecasted and selects a forecasting device from a forecasting part 3 based upon the judged result to forecast the input data. When the pattern of the input data to be forecasted is not matched with any past data pattern, the latest actual data stored in data bases 6, 7 are used as learning data for a neural network in an automatic forecasting device forming part 10 to automatically form a new forecasting part and the formed device is added to the forecasting part 3.