Research and Analysis of Predictive Models for Diabetes Mellitus Using Machine Learning: Active Learning vs. Random Sampling

Abstract

Diabetes Mellitus is one of the most significant chronic metabolic diseases characterized by persistent hyperglycemia affecting human beings in the last few decades. Currently, many research groups are working for building a predictive model to help prevent diabetes. However, the laboratory data and results are usually expensive to collect and are hard to access to. Thus, to reduce the size of the required dataset, this paper analyzes whether active learning can achieve higher effectiveness than random sampling when constructing a predictive model with fewer data to identify people at risk of having Diabetes Mellitus before diagnosis. Testing accuracy and the area under the receiver operating characteristic curve (AROC) were used to evaluate the discriminatory capability of these models. Through modeling experiments and comparisons of the two measurements, the results show that models of active learning methods can perform better than the ones using random sampling while only teaching them a small-size of the dataset.