Minimal Features based Non Invasive Cholesterol Computation using Machine Learning

Cholesterol level computation has always been an invasive and time-consuming process. This significantly hinders quick detection and delays treatment. Good and bad cholesterol exist in our body, but the latter leads to plaque formation and blocked arteries. These uncontrolled levels of bad cholesterol can cause various cardiovascular and other diseases. Conventional cholesterol calculation requires a clinical laboratory setup where specific analytical procedures need to be conducted on blood drawn invasively. As a result, a user-friendly, non-invasive method of monitoring cholesterol has become imperative. Here, the use of machine learning techniques is explored for the computation of cholesterol from the demographic and vital features of 573 subjects with ages ranging from 18 to 70. Five regressor models were implemented and compared to explore any correlation of the non-invasive features with each factor in the cholesterol lipid profile. Promising results are observed with support vector regressors and extreme gradient boosting regressors. Hence, this study can pave the way to designing and deploying a simple, reliable, non-invasive cholesterol monitoring device.