Multi-Disease Prediction Using Machine Learning: A Web-Based Diagnostic Support System for Diabetes, Heart Disease, and Parkinson\'s Disease
Multi-Disease Prediction Using Machine Learning: A Web-Based Diagnostic Support System for Diabetes, Heart Disease, and Parkinson\'s Disease
Where did the research take place?
The study site has not been established. Author addresses may differ from where the research occurred.
A plain-language reading has not been prepared for this paper yet.
Original abstract
Chronic non-communicable diseases such as diabetes mellitus, cardiovascular disease, and Parkinson’s disease are significant causes of morbidity and mortality. Screening is limited by the accessibility to medical experts and the cost of appropriate diagnostic tests. In this paper, we present a diagnostic support system based on a unified web platform that classifies patients according to the risks of developing three diseases based on regularly collected clinical or audio data using classical supervised learning algorithms. For the diabetes mellitus classification task (Pima Indians dataset, 8 features) and Parkinson’s disease detection (Oxford voice recording dataset, 22 features), Support Vector Machine classifiers with linear kernels were trained. A Logistic Regression classifier was used for predicting occurrences of heart disease (UCI Cleveland dataset, 13 features). The training results in the form of the classifiers were serialised with Pickle and implemented as a web application with the Flask framework and MySQL database software that differiates between the administrators’ and patients’ interfaces. With the stratified 80/20 holdout validation method, the system demonstrated accuracy of 77.3%, 85.2%, and 87.2% for the diabetes mellitus, heart disease, and Parkinson’s disease detection tasks, respectively. The paper describes mathematical algorithms used for machine learning in detail and presents computational specifics of the Support Vector Machine implementation for the Pima Indians dataset as an example. Additionally, the accuracy, precision, recall, and F1-score metrics were calculated based on the confusion matrices for each of the tasks.