Klasifikasi Kadar Glukosa Darah Keluaran Alat Non-invasif Menggunakan Regresi Logistik Ordinal dengan Peringkasan Luas

Abstract

Diabetes Mellitus (DM) is the silent killer because its symptoms tend to go unnoticed. Blood glucose level checks must be performed periodically to control blood glucose levels for DM and non-sufferers. The IPB Non-Invasive Biomarking Team developed a non-invasive monitoring device to check blood. The tool uses the spectroscopy principle and produces an output in the form of a residual value of light intensity. A method is needed to predict the category of blood glucose levels based on the measurement results of non-invasive tools. Classification modeling is one of the methods that can be used to analyze the relationship between the blood glucose level class of invasive measurement results and the residual value of the intensity of non-invasive measurement results. One of the commonly used classification methods is ordinal logistic regression. This method generates the final model of a cumulative opportunity logit function with an opportunity value for each class as a differentiator between classes. Light spectrum-based data used as predictor X changes often provide changes that correlate with each other. The principal component analysis reduces its dimensions to become a new set of changes that do not correlate. A good data summarization approach at the preprocessing stage is also necessary to provide good modeling. Several summarization methods have been carried out in previous studies. Graph area summation in the period is the best summarization method because it can take advantage of the general data information. This study uses the ordinal logistic regression method as a modeling method by applying principal component analysis and graph area summation applied to 2017 data and 2019 data. Classification modeling in the 2017 data had a balanced accuracy value of 64,64%. Classification modeling in the 2019 data produced a balanced accuracy value of 57,57%. The design used in the 2017 tool and the 2019 tool is different, causing the residual intensity graph of the non-invasive measurement results to be read differently. The 2017 data model is better applied to homogeneous data and the 2019 data model is better applied to heterogeneous data.
 
 Keywords: blood glucose levels, non-invasive tool, graph area summation, ordinal logistic regression, principal component
 analysis.