Privacy Protected Medical Data Classification in precision medicine using an Ontology-based Support Vector Machine in the Diabetes Management System

Diabetes is a serious health issue across the globe. As stated by the International Diabetes Foundation, currently 425 million people live with diabetes globally, and another 300 million people are expecting to be at higher risk of diabetes in the year 2030.Hence, there is an urgent clinical need for an early prognosis, diagnosis, and management of diabetes and its complications. In this context, numerous intelligent machine learning and data mining approaches including Support vector machine (SVM) have been exploited for diabetes management.SVM is a prevailing supervised and discriminative data classifier which assists the healthcare and biomedical professionals to ascertain unknown data patterns by training a large volume of real-time data. Since this database is the private asset, the sensitive information should be safeguarded without compromising the utility. Even though SVM is a fast and accurate machine learning technique, it does not have the capacity to represent semantically the classification and reasoning rules which can enable more precise classification. Therefore, the objective of this research is three-fold. Firstly, in order to protect sensitive clinical data, we introduce the Kronecker product and Crow Search Algorithm (CSA) based coefficient generation technique. Secondly, we design diabetes ontology to define domain concepts and relationships and allow medical datamining. Finally, we implement the Ontology-based Support Vector Machine (Ont-SVM) classifier by assimilating privacy protection, ontology, and SVM-based classifier for the diagnosis of diabetes.The experimental results on a real-world dataset, the Pima Indian database at the UCI repository, illustrate that the proposed Ont-SVM outperforms other existing approaches in terms of privacy, accuracy, specificity, and sensitivity.