Multi-label Chronic Non-Communicable Diseases Prediction Using Machine Learning

Abstract

Chronic non-communicable diseases are becoming more prevalent every day, especially in developing countries like Ethiopia. These disorders must be treated as soon as possible. Machine learning-based models can be used to predict chronic diseases before medical diagnosis. Numerous prediction models have been used for a single disease. However, they have overlooked the suffering of many people from multiple related diseases that have a comparable effect due to physiological irregularities. So far, there is a paucity of information regarding multi-label predictive models for hypertension and diabetes. Therefore, this study aimed to design and develop a multi-label predictive machine learning model and application which predicts diabetes and hypertension simultaneously. The study used a physical examination dataset from the Ethiopian Public Health Institute, of which 30% are for testing and 70% are for training. The linked diseases were predicted using common risk variables and label correlation. A multi-label feature selection was performed using a chi-square test, correlation-based, combing fisher's-score with chi-score, and using the featurewiz python library. Multi-label synthetic minority oversampling technique is used to handle the class imbalance problem. Moreover, problem transformation approaches, binary relevance, and classifier chain were integrated with five common machin