Abstract:
Introduction: The number of people affected by hypertension is staggering, with an estimated one billion people living with the disease worldwide. It has been shown that machine learning (ML) models surpass clinical risk; nevertheless, there isn't much research using ML to predict hypertension in Malaysia.
Materials and methods: A study is being conducted using ML analyses to predict hypertension using secondary data from population-based surveys, such as the National Health & Morbidity Survey (NHMS) 2015. The dependent or target variable was hypertension status and 24 features. Three standard ML-based classifiers, which are logistic regression (LR), decision tree (DT) and artificial neural network (ANN), were used to predict hypertension and the associated factors that influence hypertension were obtained from filter-based feature selection, which are feature weight by information gain, feature weight by information gain ratio and feature weight by correlation. Results: Out of 11,520 respondents, 4,175 (36.24%) adults had hypertension. LR is the best model to predict hypertension since LR has the highest accuracy (76.73%) compared to DT and ANN (73.02%). In terms of odd ratio explanation, a person who does not have diabetes mellitus is 2.05 odds likely to have hypertension, and a person who does not have hypercholesterol has 1.67 odds of having hypertension, and with an increase in the age of adults, 6.0% are less likely to have hypertension.
Conclusion: From LR model, the essential features that influence hypertension in adults were diabetes mellitus, hypercholesterolemia status, age, waist circumference, marital status, occupation, education, and total household income.
