A Machine Learning Model to Identify Insulin Resistance in Humans

Abstract

T2DM is a large challenge because it's predicted to affect 693 million people by 2045. There is currently no simple or non-invasive method to measure and quantify insulin resistance. Following the release of non-invasive devices that track glucose levels, one might be able to identify insulin resistance without having to use invasive medical tests. In this work, insulin resistance is recognized based on non-invasive techniques. Eighteen parameters are used to identify a person with a high likelihood of insulin resistance: consisting of age, gender, waist size, height, etc., and an aggregate of those parameters. Each output of a function choices technique is modeled using a range of algorithms, including logistic regression, CARTs, SVM, LDA, KNN, etc on CALERIE study dataset and the findings are verified over stratified cross-validation. And in comparison, to 66% Bernardini et al & Stawiski et al, 61% Zheng et al, and 83% Farran et al, the accuracy of different variations for the identification of insulin resistance. Another advantage of the proposed approach is that an individual can also predict insulin resistance daily, which in turn will allow physicians to monitor diabetes risk more accurately. While the identical isn't always almost feasible with medical procedures.