Managing Diabetes Therapy through Datastream Mining

In insulin-dependent diabetes mellitus (IDDM) therapy, a suitable insulin dosage taken at the appropriate times is needed for each patient to sustain the necessary blood-glucose level for his or her body. In this article, a datastream mining approach is proposed that can computationally derive real-time decision rules for formulating IDDM therapy based on insulin prescription records and patients' blood-glucose reactions. Decision rules are based on the latest health conditions, which are monitored continuously from the patient rather than from a historical data archive of a population accumulated over years. Hence, the rules are adaptive and more accurately predict whether a medical implication will occur, given that glucose levels fluctuate under different medical effects, such as lifestyle changes, medication type, or other external factors. A computer simulation experiment is conducted for evaluating the most suitable datastream algorithms with respect to accuracy and speed.