An End-to-End Diabetes Diagnosis System Powered by Machine Learning

Abstract

The onset of providing medical care in the traditional healthcare model of reactive medicine begins with the patient. It’s fairly common for patients to become ill and choose not to visit their doctor, unless they feel there’s sufficient reason to do so. This is problematic because when diseases are diagnosed in later stages, the chances of successful treatment and even survival in some cases are dramatically reduced. The future of healthcare is the antithesis of this model; it is a proactive one where late-stage diagnosis is rare. In this thesis, I explore this future by building a 24/7, non-invasive companion doctor; a platform intended to perform real-time disease diagnosis and monitoring powered by various machine learning methods. In particular, I focus on two disease categories, type-1 diabetes and type-2 diabetes, but the methodologies and results were constructed in such a way that they would be easily generalizable to other disease categories.