Anomaly Detection using Transition Graphs on a Longitudinal Diabetes Dataset

Type II Diabetes Mellitus is rapidly becoming the most prevalent disease in modern society. Anomaly detection is an important tool for addressing the disease, as new discoveries are often found when studying abnormal cases. This thesis explores a graph theoretical approach for detecting anomalies in a longitudinal diabetes dataset. The transition graph model proposed in this study bypasses the limitations encountered by dynamic graph anomaly detection methods when attempting to detect anomalies in longitudinal datasets with as few as two time points.