PERFORMANCE OF LASSO AND ELASTIC-NET METHODS ON NON-INVASIVE BLOOD GLUCOSE MEASUREMENT CALIBRATION MODELING

Abstract

Diabetes Mellitus (DM) is a disease that can occur in humans caused by conditions of high blood glucose levels (hyperglycemia). Detection of blood glucose levels can be done using invasive methods (injuring) and non- invasive methods (with infrared rays). Analytical methods are needed to model these results to obtain estimates of blood glucose levels. An alternative approach that can be used to analyze the relationship between invasive and non- invasive blood glucose levels is the calibration model. Problems that often occur in calibration modelling are multicollinearity and outliers. These problems can be overcome by adding new data, applying principal component analysis, and using LASSO and Elastic-Net regression to overcome calibration problems. The research data used was invasive and non-invasive blood glucose data in 2019, with as many as 74 respondents. The results of the study concluded that the summarization of the trapezoidal area in calibration modelling provides a good estimate. The performance of the Elastic Net method provides better prediction results than other models, with an RMSE value of 22.39. It has the most significant positive correlation value of 0.97, which means close to 1 so that the performance of the Elastic Net method can handle calibration modelling.