Machine learning application for classifying serum uric acid level with photoplethysmogram

Cardiovascular disease (CVD) is a cause of 17.9 million deaths each year globally. Among the risk factors of CVD are hypertension, hyperlipidemia, hyperglycemia, and hyperuricemia. In hyperuricemia, serum uric acid (SUA) level can be detected only with a blood test, a microinvasive procedure that is costly and incur minimal pain and requires patients to travel to a hospital for regular monitoring. In this paper, a machine learning technique is proposed for SUA level classification. A total of 74 subjects participated in this study, and 20 features were extracted from the pre-processed photoplethysmogram (PPG) morphology. Blood tests were performed, and the results were used as inputs and labeled accordingly. Results show that based on p-value, the onset-diastolic width of the right and onset-systolic width of the left hand are the most significant features for SUA classification. Artificial neural network has an accuracy of 86.36%, which is higher than that of three machine learning methods: k-nearest neighbor, support vector machine, and decision tree.