INFERENCE METHOD BASED ON SIMILARITY OF EXAMPLES

摘要

PROBLEM TO BE SOLVED: To solve the problems in the conventional technology that the accuracy of inference may be reduced when the distribution of examples is not uniform and a function for calculating the confidence of inference is not effectively operated. SOLUTION: After executing a step for calculating the similarity of examples, a similar example is determined while utilizing the distribution information of field values for calculating the inference values of examples having high similarity. Since the similar example is determined by considering the peripheral distribution of unknown examples, a suitable similar example can be selected and inferred, even when the distribution of examples is not uniform. Since the confidence is determined, while utilizing the distribution information of field values for calculating inference values of examples having high similarity, the suitable confidence can be calculated, even when the distribution of examples is not uniform.