A search method of searching for a feature that affects an output result of a machine learning model, the search method includes: a first step of applying, to all training data, at least one or more separate filters combining at least one or more feature presence/absence determining devices that determine presence or absence of a feature on a plurality of sets of correct answer data that is positive and correct answer data that is negative and information on whether the pieces of the data is positive; a second step of applying the pieces of training data generated in the first step to separate machine learning to separately execute machine learning; and a third step of outputting information that extracts a new feature using a verification result obtained by inputting verification data to separate machine learning after the machine learning.