Machine Learning to identify factors that affect Human Systolic Blood Pressure

This paper employs different machine learning algorithms to perform a regression study to predict systolic blood pressure (SBP) levels. We used blood pressure dataset of Dr. Raymond Lam, GlaxoSmithKline, Toronto, Ontario, Canada in this study. There are 500 patients in the dataset, 250 have normal blood pressure level and the other 250 have hypertension. There are 500 predictors in the dataset. 17 predictors are patients' non-genomic information and the rest are 483 genetic markers. In this paper, we have selected only the following 13 factors as predictors in this study to reduce the complexity of the problem. The predictors included in this study are 'gender', 'married', 'smoke', 'exercise level', 'age', 'weight', 'height', 'alcohol consumption', 'treatment for hypertension', 'stress level', 'salt intake level', 'income' and 'education level'. The regression model that gave the lowest root mean square error in SBP of 25.68 for the 13 predictors is Gaussian Process Regression using the squared exponential function in the regression model. Although the RMS value was quite high, it was sufficient to draw some conclusions and identify the factors that do affect the SBP level.