A Machine Learning Based Approach for the Identification of Insulin Resistance with Non-Invasive Parameters using Homa-IR

Type-2 diabetes mellitus (T2DM) is a significant concern since it is anticipated to reach over 693 million individuals by 2045. Identification and quantification of insulin resistance requires a particular blood test which is complicated, time-consuming and most importantly invasive, making it not feasible for routine day-to-day activities of a human being. With the development of recent machine learning approaches, identification of insulin resistance could be performed without clinical procedures. In this work insulin resistance isidentified based on machine learning approaches using non-invasive techniques. Eighteen parameters are used for identification of insulin resistance; such as age, gender, waist size, height, etc. and a combination of these parameters. Experiments are conducted on the CALERIE dataset. Each output of the feature selection method is modelled over different calculations such as logistic regression, CARTs, SVM, LDA, KNN, etc. The proposed approach is verified using Stratified cross-validation test. Results show that using Logistic regression, SVM and few other versions for identification of insulin resistance, accuracy up to 97% has been achieved with standard deviation of 1% compared to 66% with Bernardini et al. [5] & Stawiski et al. [11], 61% Zheng et al. [35] and 83% Farran et al. [32]. Major benefit of the proposed approach is that a person may forecast the insulin resistance and thus future odds of diabetes may be monitored on a daily basis using non-clinical approaches. While the same is not practically possible with clinical procedures.