Predicting insulin dosage for diabetic patients to reach optimal glucose levels. (c2012)

Abstract

Endocrinologists treating diabetes mellitus find it overwhelming to keep track of the enormous amount of data gathered from the multiple measurements retrieved daily by their patients, throughout the years. This makes it hard for them to find a certain pattern in the data that could help them assign, to each patient, the optimal insulin dosage. This forces them to seek trial and error until they find the individualized insulin doses, required by each patient, to reach their optimal glucose levels. Hence, there is a great read to automate the process of estimating the glucose level. For this, we propose two machine learning techniques and one heuristic. In particular, we present C4.5, Case-Based Reasoning and genetic algorithms. We validate our approach on a data set obtained from the UCMI online machine learning repository. Obtained results are promising. Case-Based Reasoning outperformed both C4.5 and Genetic Algorithms.