En-RfRsK: An ensemble machine learning technique for prognostication of diabetes mellitus

Abstract

In human bodies, diabetes mellitus is a serious illness catalyzed by high glucose levels. In the event that diabetes remains untreated, it can lead to other basic health issues. This study examines different human body attributes in an effort to prognosticate diabetes. An ensemble approach named, En-RfRsK is proposed to predict the risk of diabetes mellitus which is a voting classifier that encompasses three machine learning techniques, viz., random forest (RF), radial support vector machine (R-SVM), and K-nearest neighbour (KNN). RF utilizes the performance of a large number of relatively unequal models or trees. R-SVM uses a function that changes in value depending upon how far it is from the origin. KNN learns the non-linear decision boundaries of the diabetes data. The advantages of these machine learning (ML) techniques have been exploited in the proposed approach. The proposed approach is an ensemble of existing ML algorithms, since ensemble algorithms are considered to be more accurate and flexible than single classifiers. With its better accuracy and predictive capabilities, it delivers the best solutions. Experiments were administered using PIMA diabetes dataset. It is evident from the experiments that the proposed approach outperforms the existing base classifiers and state-of-the-art machine learning diabetes mellitus prediction systems. The accuracy obtained by the En-RfRsK approach is 88.89%.

Keywords

• Diabetes mellitus
• Ensemble
• Machine learning
• Voting classifier