calibration modeling in non invasive blood glucose levels using support vector regression

Abstract

accurate measurement of blood glucose levels is needed in the treatment and prevention of diabetes mellitus. blood glucose levels can be measured by injuring (invasive) and not injuring (non-invasive) parts of the body. invasive measurements can cause discomfort for patients and require relatively more expensive costs. one alternative to overcome this problem is to develop a non-invasive measurement tool. the relationship between the two measurement results can be modeled using calibration. the aim of this study was to predict non-invasive blood glucose levels. the data used were part of the data on prototype clinical trial and development research for monitoring tools for non invasive blood glucose levels at the bogor agricultural university (ipb). the approach method used was support vector regression (svr) for high dimensional data in the calibration model. the results indicated that the svr using a base radial kernel was the best model. prediction results of non-invasive blood glucose levels had closer blood glucose levels to the results of invasive measurements. this wass supported by a greater value of the coefficient of determination and the smaller value of root mean square error prediction. furthermore, it can be concluded that the model obtained could be used to predict non-invasive glucose levels and could be recommended to related sectors. however, these results were still in a narrow range of data so that it becomes a suggestion for related parties to use more samples in order to widened the range of data.