Learning method of linear discriminant function

摘要

PURPOSE:To provide an optimum discriminant function used for the diagnostic rule of a diagnostic system by means of a linear plan method. CONSTITUTION:A positive/negative example separation part 11 divides an inputted diagnostic example 10 into a positive example 12 and a negative example 13. A positive example conversion part 14 generates a linear limited expression 16 obtained by adding a first variable reducing a constant item to the positive example and making an unequal sign equal. A negative example conversion part 15 generates a linear limited expression 17 obtained by adding a second variable increasing the constant item to the negative example and encoding the unequal sign. A linear plan execution part 18 sets the total of the first variable and the second variable to be a target function 19, and generates a discrimination function 20 so that the target function 19 becomes a minimum with the limited expressions 16 and 17 against the positive and negative examples as limited conditions. The diagnostic system generates the whole discriminant function by repeatedly using a learning system 1 for plural illnesses.