A remote healthcare monitoring framework for diabetes prediction using machine learning

Abstract

Diabetes is a metabolic disease that affects millions of people each year. It is associated with an increased likelihood of vital organ failures and decreased quality of life. Early detection and regular monitoring are crucial for managing diabetes. Remote patient monitoring can facilitate effective intervention and treatment paradigms using current technology. This work proposes an end-to-end remote monitoring framework for automated diabetes risk prediction and management, using personal health devices, smart wearables and smartphones. A support vector machine was developed for diabetes risk prediction using the Pima Indian Diabetes Database, after feature scaling, imputation, selection and augmentation. This work achieved the performance metrics of accuracy, sensitivity and specificity scores at 83.20%, 87.20% and 79% respectively through the tenfold stratified cross validation method, which is competitive with existing methods. Patients can use multiple healthcare devices, smartphones and smartwatches to measure vital parameters, curb the progression of diabetes and close the communication loop with medical professionals. The proposed framework enables medical professionals to make informed decisions based on the latest diabetes risk predictions and lifestyle insights while attaining unobtrusiveness, reduced cost, and vendor interoperability. INTRODUCTIONDiabetes is a metabolic disorder that causes a lack or resistance to insulin, which is a hormone critical for the regulation of blood sugar levels. The common symptoms are polyuria, polydipsia, rapid weight loss, vision blurriness and fatigue [1]. This leads to severe complications like strokes, blindness, miscarriages and organ failures. The World Health Organization in 2014 reported that approximately 422 million people worldwide have diabetes, and estimates 1.6 million deaths directly caused by the disease [2]. Medical guidelines recommend the need for early diagnosis to identify risk-prone people and for patients to proactively self-monitor their lifestyle to mitigate risk factors [3]. Remote patient monitoring (RPM) can help prevent the staggering number of deaths caused by diabetes by means of early detection and timely warnings to the patient and medical professionals. RPM reduces the need for regular checkups, gauge continuous treatment effectiveness and allow intervention techniquesThis is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited.

A Comprehensive Review of Various Diabetic Prediction Models: A Literature Survey

“…Features were selected using logistic regression methods, and the random forest classifier gave the optimum result. Ramesh et al [ 38 ] proposed a framework for automating the process of prediction of diabetes and made use of health devices such as smartwatches or a smartphone. The machine learning classifier used by the authors to classify the dataset is a support vector machine.…”

Towards Validating the Effectiveness of Obstructive Sleep Apnea Classification from Electronic Health Records Using Machine Learning

“…The traditional machine learning models of k-Nearest Neighbours (kNN), Support Vector (SVM) Machines and Logistic Regression (LR) are used as baseline to benchmark the performance of the ensemble techniques [50]. KNN is non-parametric learning algorithm which distributes similar instances in the same proximity defined by the Euclidean distance, and classifies new unknown instances by majority vote of their k nearest instance neighbours.…”

Machine learning models for classification and identification of significant attributes to detect type 2 diabetes

“…Ismail et al [ 22 ] provided a taxonomy of significant factors where different machine learning algorithms were used with or without feature selection processing. In addition, Ramesh et al [ 44 ] implemented multivariate imputation by chained equations (MICE) method for handling missing values of primary diabetes dataset. Subsequently three feature selections (chi-squared test, extremely randomized trees, and least absolute shrinkage and selection operator (LASSO)) and some classifiers such as KNN, LR, Gaussian NB, and SVM were used to investigate this dataset.…”