Design and Deployment of E-Health System in Perspective of Developing Countries: Machine Learning Based Approach (Preprint)
BACKGROUND We are living in a world where data science and machine learning is tightening its grasp on many sectors of modern life. The medical sector is not an exception. In developing countries, healthcare is one of the domains that need immediate attention. Due to the lack of manpower and technical resources, a large number of people in these regions do not receive proper medical care. Designing an E-health system with the help of machine learning and web technologies would be a great aid in such circumstances. OBJECTIVE This proposed E-health System will assist the medical professionals in determining diseases. Moreover, the system will be also helpful for the patients to check whether they have been diagnosed correctly. Based on their diagnosis results they can get medical specialist recommendation and medicine suggestions from the system. The automation of identifying the diseases and suggestion models with the help of machine learning will be cost-efficient and time-saving compared to the traditional methods. The main objective of this E-health system is to provide health care with the help of sustainable and realistic machine learning technologies. METHODS In this research, for the disease identification part, machine learning techniques have been applied to identify three diseases which are Dengue, Diabetes, and Thyroid. Decision Tree, Gaussian Naive-Bayes, Random Forest, Logistic Regression, k-Nearest Neighbors, Multilayer Perceptron, and Support Vector Machine Classifiers have been used for all three diseases. The E-health system comprised of disease identification model, medical specialist recommendation model, and the medicine suggestion model has been deployed on the web. The medical specialist recommendation model and the medicine suggestion model results are based on the finding of the disease identification model. Any user can insert their disease-specific data to use these three features of the E-health system. RESULTS For the disease identification model, Multilayer Perceptron for Dengue, Logistic Regression for Diabetes, and Random Forest for Thyroid performed the best with accuracies of 88.3%, 82.5%, and 98.5% respectively. These classifiers also showed good precision, recall, and F1 score. CONCLUSIONS The E-health system has performed well with real-time data. By making the dataset more enriched, the disease identification model will be more robust and thorough. Moreover, usability and acceptance tests can help us in finding different real-time scenarios of the E-health system.