DEVELOPMENT OF MACHINE LEARNING ALGORITHM FOR VALUE CONVERSION AND SYNTHETIC DATA GENERATION IN NON-INVASIVE GLUCOSE MONITORING DEVICES USING NEAR INFRARED SPECTROSCOPY METHOD

• Main Author: Rahmi Fakhrunnisa, Iftika
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/50807
• Institution: Institut Teknologi Bandung

Diabetes mellitus is a metabolic disease caused by an imbalance in blood glucose levels with the hormone insulin. Based on the Basic Health Research conducted by the Ministry of Health of the Republic of Indonesia, the prevalence value increased by 4%. Due to the high prevalence value in Indonesia, diabetes mellitus is a dangerous disease that is not only life threatening but will have an impact on the economy due to health costs. Therefore, sufferers are required to perform healing actions and control blood sugar levels regularly. Most of the devices used to perform this evaluation on the market still use an invasive method whose use can cause discomfort to patients. Therefore, we need a blood sugar measuring device that is safer, more comfortable, practical for sufferers, namely a noninvasive method of diagnosis. Near Infrared Spectroscopy is a molecular concentration measurement method that releases the resulting absorbance value which will be converted into blood sugar levels (mg / dL). Value conversion is done using machine learning as a computational technique that is better than the statistical approach. Due to the COVID-19 outbreak, the process of this final project will focus on developing value conversions and increasing blood sugar reading ranges. It is important to extend the reading range of the tool so that the tool can work on data of greater value. Synthesis data was made using the partially synthetic method from the dataset obtained from Galih, 2019. From the development of value conversion and synthesis of synthetic data, the mean absolute error of synthetic data is 3.87 mg / dL in the reading range of 67-188 mg / dL. To make it easier for patients to evaluate blood sugar levels, the device will be connected to a smartphone and the operating system used is Android this is due to the large number of users on the mobile operating system, which increases the likelihood of using a non-invasive method of glucose monitoring device in near infrared spectroscopy .