Decentralized Learning with Virtual Patients for Medical Diagnosis of Diabetes

Abstract Machine learning, applied to medical data, can uncover new knowledge and support practices. However, analyzing data by machine learning methods presents a trade-off between accuracy privacy. To overcome the trade-off, we apply collaboration analysis method data. This using artificial dummy enables compare distributed information without original The purpose of our experiment is identify patients diagnosed with diabetes mellitus (DM), 29,802 instances real obtained from University Tsukuba Hospital 01/03/2013 30/09/2018. whole divided into number datasets simulate different hospitals. We propose following improvements for analysis. (1) Making which has reality (2) non-linear reconverting functions comparable space. Both be realized generative adversarial network (GAN) Node2Vec, respectively. improvement effects GAN scores more than 10% over random numbers. Furthermore, effect re-conversion Node2Vec anchor about 20% higher linear Our results reveal that appropriate modifications, depending on type, improves performance.