A discrimination device and a machine learning method are provided with which it is possible to detect errors in annotation. This discrimination device 1 comprises: a sub-data set extraction unit 34 that extracts, from a plurality of learning data to which labels have been applied, a sub-learning data set used for learning, and a sub-verification data set used for verification; a learning unit 110 that performs supervised learning based on the sub-learning data set, and generates a learned model for discriminating among labels based on data according to the object; a discrimination unit 120 that performs a discrimination process using the learned model for each of the learning data items included in the sub-verification data set; a verification result recording unit 40 that associates the result of the discrimination process with the learning data and stores the resu and an error detection unit 42 that detects learning data for which the applied label may be wrong, on the basis of the result of the recorded discrimination process that was associated with each learning data item.
展开▼
