Performance Investigation of Principal Component Analysis for Diabetic Detection Using SVM with Kernel Functions

Diabetes mellitus has been increasing in a day-to-day scenario where this deadly disease increases the blood sugar level. In earlier days, only very few people were affected by this disease. In the present years, our immobilized activity, unhealthy eating, and mental stress lead to millions of adults who are mostly affected by this. In some instances, even the existence of this disease is unseen or unknown until they undergo a medical examination. Before all these things could have happened, the disease had sharpened itself by thriving in society; even some people ended up dying at very early ages. If we fail to diagnose, it leads to severe complications. This paper aims to create a model that can detect diabetes early to avoid the risk of diabetes. The proposed is an investigation model that selects the features of the diabetes dataset using a principal component analysis (PCA) and support vector machine (SVM) to classify the selected features. In this, SVM has been analyzed with different support vector kernels such as linear, polynomial, RBF, and sigmoid. Experimentation has been carried out in PIMA Indian diabetes dataset and the diabetes dataset to evaluate the performances of various principal components with SVM kernels. In the experimentation, it has been analyzed that the principal components 1 and 2 with SVM RBF will be the best suitable for predictive modeling of diabetes detection.