RANCANG BANGUN PENDETEKSI KADAR GULA DARAH NON-INVASIF PORTABEL BERBASIS IOT

DESIGN AND DEVELOPMENT OF A PORTABLE NON-INVASIVE BLOOD GLUCOSE LEVEL DETECTOR BASED ON IOT 

Diabetes Melitus merupakan masalah kesehatan global yang memerlukan pemantauan kadar gula darah secara rutin. Metode invasif yang umum digunakan saat ini menimbulkan ketidaknyamanan dan risiko infeksi. Penelitian ini bertujuan untuk merancang dan membangun prototipe alat pendeteksi kadar gula darah non-invasif portabel berbasis Internet of Things (IoT) menggunakan mikrokontroler ESP32 dan sensor Photoplethysmography (PPG) MAX30100. Sistem ini bekerja dengan mengukur fluktuasi sinyal PPG (Inframerah dan Merah) dari ujung jari, yang kemudian diolah untuk mengekstraksi fitur fitur seperti detak jantung (BPM), saturasi oksigen (SpO2), rasio AC/DC IR, dan rasio AC/DC Red. Fitur-fitur ini, bersama dengan data usia subjek, digunakan dalam model regresi linear multivariat untuk mengestimasi kadar glukosa darah. Data hasil pengukuran ditampilkan pada LCD dan disimpan secara lokal pada modul SD Card, serta dikirim secara real-time ke Google Sheets melalui konektivitas WiFi. Pengujian validasi dilakukan pada 31 sampel data, menunjukkan korelasi statistik yang signifikan antara fitur PPG dan kadar glukosa darah aktual. Prototipe ini mencapai tingkat akurasi total sebesar 92% dengan rata-rata persentase kesalahan absolut (MAPE) sebesar 8%. Performa terbaik tercatat pada mode pengukuran "Sewaktu" dan "Setelah Makan" dengan rata-rata kesalahan 5%. Namun, akurasi sedikit menurun pada mode "Puasa" (rata-rata kesalahan 14%), terutama pada subjek dewasa di atas 40 tahun, yang kemungkinan disebabkan oleh penurunan elastisitas pembuluh darah yang memengaruhi kualitas sinyal PPG. Penelitian ini berhasil membuktikan kelayakan konsep (proof-of-concept) alat pendeteksi gula darah non-invasif yang nyaman dan portabel. Meskipun demikian, disarankan untuk melakukan validasi model lanjutan dengan ukuran sampel yang lebih besar dan beragam, serta eksplorasi model machine learning non-linear untuk meningkatkan generalisasi dan ketahanan terhadap variasi sinyal fisiologis. 

Diabetes Mellitus is a global health issue requiring routine blood glucose monitoring. Current common invasive methods cause discomfort and infection risks. This research aims to design and develop a prototype of a portable non-invasive blood glucose level detector based on the Internet of Things (IoT) using an ESP32 microcontroller and a MAX30100 Photoplethysmography (PPG) sensor. The system operates by measuring PPG signal fluctuations (Infrared and Red) from the fingertip, which are then processed to extract features such as heart rate (BPM), oxygen saturation (SpO2), AC/DC IR ratio, and AC/DC Red ratio. These features, along with the subject's age data, are used in a multivariate linear regression model to estimate blood glucose levels. Measurement results are displayed on an LCD and stored locally on an SD Card module, as well as sent in real-time to Google Sheets via WiFi connectivity. Validation tests were conducted on 31 data samples, demonstrating a significant statistical correlation between the extracted PPG features and actual blood glucose levels. The prototype achieved a total accuracy of 92% with a Mean Absolute Percentage Error (MAPE) of 8%. The best performance was observed in "Random" and "Post-meal" measurement modes, with an average error of 5%. However, accuracy slightly decreased in "Fasting" mode (average error 14%), particularly in subjects over 40 years old, likely due to reduced vascular elasticity affecting PPG signal quality. This research successfully demonstrates the proof-of-concept for a comfortable and portable non-invasive blood glucose detector. Nevertheless, further model validation with larger and more diverse sample sizes, as well as exploration of non-linear machine learning models, are recommended to improve generalization and robustness against physiological signal variations.