PURPOSE: To exactly perform inference at all times by always easily and convergently learning proper diagnostic knowledge considering frequency again by successively correcting relation degree data so that the relation degree data of a true factor can be enlarged and the relation degree data of factors excepting for this true factor can be reduced each time this true factor is discriminated.

CONSTITUTION: The relation degree data are successively corrected so that a relation degree Wij of a true factor (t) can be enlarged and a relation degree Wij of factors (j) excepting for this factor (t) can be reduced each time this factor (t) is discriminated. In a step 101, a present abnormal degree Ii of a phenomenon (i) is inputted. In 102, the true factor (t) is inputted and each time this true factor is discriminated, a relation degree correction amount dWit is calculated by using a maximum diving method with an error E as an evaluation function. In the 102, on the other hand, the relation degree Wij is corrected by calculating a relation degree correction amount dWij- of factors (j) excepting for the true factor (t). In 103, a degree Oj of certainty is calculated. In 104, the diagnostic result is displayed.