Diabetes Prediction: A Comparison Between Generalized Linear Model and Machine Learning

Recently, many chronic diseases have directly impacted human health. Many diseases are rampant and cause significant damage to humankind. Technological advancement has proven that diseases can be cured early. However, some diseases cannot be cured entirely but can be prevented. One of them being diabetes. If diabetes remains untreated and undiagnosed, many complications arise. The tedious diagnostic process leads the patient to visit a diagnostic center and see a doctor. The growth of technological solutions is solving this vital problem. The study’s goal is to forecast whether a patient would get diabetes based on eight input variables: pregnancy, glucose, skin thickness (ST), blood pressure (BP), insulin, BMI, diabetic pedigree function (DPF), and age are all factors to consider. The output variable of the study is named “outcome,” which is a binary variable taking values 1 or 0. The number 1 represents the existence of diabetes, whereas the value 0 represents the absence of diabetes. This chapter compares the results of classification algorithms, viz., binary logistic regression model and a support vector machine to detect diabetes early. The models are evaluated using the accuracy score metric to find the best model. The data used for the study is taken from GitHub.