AI-Based Methods for Predicting Required Insulin Doses for Diabetic Patients

Abstract

Treating diabetes mellitus requires patients to retrieve multiple measurements daily over multiple years. This results in an enormous amount of data. Endocrinologists need to find a certain pattern in this data that would help them determine the optimal dosage of insulin to administer to each patient. However, keeping track of the data for this purpose is overwhelming. As a result, they often follow a trial and error approach until they find the individualized insulin dosage, required for each patient, to reach their optimal glucose level. Hence, there is a great need to automate this process. In this paper, we propose and compare three techniques two of which are Artificial Intelligence techniques, namely C4.5 and Case-Based Reasoning, and the third one is a meta-heuristic namely genetic algorithms. The performance of the three algorithms is evaluated on a data set found in the public UCMI repository.