Discovering Symptom Subgroups via Representation Learning and Clustering

Syndrome is a group of co-occurring symptoms that characterize a particular physical abnormality, and symptom subgroups can be used for the diagnosis and treatment of diseases. Discovering symptom subgroups from medical records set is usually faced with the challenge of selecting a proper data representation. In this work, an algorithm of discovering symptom subgroups based on representation learning and clustering of symptoms is proposed. To obtain a suitable representation of symptoms, word embedding method is adopted to avoid some issues caused by ordinal encoding and one-hot encoding methods. Moreover, to discover symptom subgroups, Fuzzy C-Means (FCM) clustering algorithm is used for symptom clustering based on the characteristics of such dataset. Finally, the effectiveness of the proposed algorithm is demonstrated on a real medical records dataset.