Investigations on novel optimization techniques for diabetes disease diagnosis and prediction using machine learning approaches

Abstract

Efficient disease management and accurate clinical diagnosis are newlinetwo vital issues in the medical industry and have a positive impact on the newlinepublic healthcare system. Diabetes Mellitus (DM) is a typical metabolic newlinedisease in which individuals struggle with high blood sugar (i.e., chronic newlinehyperglycemia). It affects most of the body parts such as the heart, kidney, newlineeyes, foot, skin, etc. Several competent Diabetes Diagnosis Systems (DDS) newlineexploit different Machine Learning (ML) algorithms for gaining valuable newlineinsights from the clinical datasets for DDS and disease management. newlineHowever, trapping into local optimum solution, lack of privacy, missing newlinevalues in the input dataset, and deficiency of incremental classification are newlinemajor issues related to conventional ML-based diabetes classification newlinealgorithms.The main goal of this work is to develop an effective DM newlineclassification model that can reliably identify patient data as normal or newlinediabetic. To reach this goal, this work suggests: (i) a hybrid optimizer-based newlineSupport Vector Machine (SVM) using Crow Search Algorithm (CSA) and newlineBinary Grey Wolf Optimization (BGWO) algorithm; (ii) a Hybrid Online newlineModel for Early Detection of diabetes disease (HOMED) by applying an newlineimproved incremental SVM (ISVM) and an Adaptive Principal Component newlineAnalysis (APCA) algorithm; and (iii) an Ontology-based SVM (Ont-SVM) newlineusing Kronecker product and CSA-based parameter optimization techniques. newlineIn the first work, a hybrid optimizer using SVM is proposed to develop an newlineeffective diabetes detection model. It assimilates a CSA and BGWO for newlineexploiting the entire capacity of SVM for detecting diabetes. newlineThe second work proposes an online diabetes classifier, called newlineHOMED by applying an APCA algorithm for missing value imputation, newlineclustering, and feature extraction and an ISVM algorithm for classifying newlinediabetes mellitus. Additionally, this work targets to develop a secured newlinediabetes classification model where the privacy of sensitive user data is newlineprotected. Hence, this work develops a secured DDS to protect newlineuser-information using Kronecker product-based CSA, which optimizes the newlineparameter of Secrecy Usage (SU). Then, the domain ontology is developed to newlinedetect DM by determining clinical resemblances between the ontological newlinerules with the trained dataset. Hence, this research implements the Ont-SVMbased newlineclassification model by integrating privacy protection, ontology, and newlineSVM-based classification algorithm as third work newline newline