Diabetes Mellitus Prediction System Using Hybrid KPCA-GA-SVM Feature Selection Techniques

Abstract

Diabetes mellitus is a serious health issue in healthcare industry, which is a type of uncontrolled level of sugar. It is a chronic disease happened to the person who are having low insulin production and increase level of blood glucose because glucose is not properly utilized by body. In the medical field, predicting the correct diabetes is an important area that is under research to define a good predictive system to help the doctors to diagnose the disease. In the predictive system, feature selection plays on vital role to select the relevant feature for classification. There are several algorithms were applied on classification of diabetes data. In this proposed work, the features are transformed into high dimensional space before selection. So that the transformation of the features will give the better selection of attributes. With this effort, the proposed work implements the Kernel Principal Component Analysis for dimensionality reduction. KPCA will reduce the features space better than PCA. Once the features are transformed, the proposed work uses Genetic Algorithm to select the relevant and optimal features from the dataset. Then at the last Support Vector Machine is used as a classifier to classify the diabetes mellitus data. The proposed research on applying feature reduction before feature selection will reduce the irrelevant features that will improve the accuracy of the classification based on the selected relevant features. This proposed algorithm on diabetes mellitus data will compare with the existing algorithms to prove the effectives of the algorithm.