Model Kalibrasi untuk Prediksi Kadar Gula Darah Non-Invasif menggunakan Regresi Kuantil Komponen Utama
Date
2021-02-23Author
Novia, Siti Arita
Erfiani, Erfiani
Wigena, Aji Hamim
Metadata
Show full item recordAbstract
Diabetes mellitus (DM), a chronic metabolic disorder, is caused by pancreas
inability to produce enough insulin (a hormone that regulates blood glucose) or of
which the body cannot use the insulin produced effectively. As a result, there is an
increase in the concentration of glucose in the blood (hyperglycemia). If it is not
immediately prevented or addressed, DM will induce complications leading to
death. DM is one of the global health problems with the fastest-growing emergency
in the 21st century, so it is necessary to take preventive actions. Including checking
blood glucose levels regularly. Furthermore, blood glucose level test is usually
conducted in invasive ways such as by using injection and a glucometer which may
injure the body. This method is takes times and spends a big expense. Based on
those phenomena, the IPB bio marking team was motivated to develop a noninvasive blood glucose monitoring device that is non-injurious for the body. The
development of this blood test requires calibration analysis model. The previous
research had shown that there were outlier data on the results of invasive blood
glucose levels measurement. However, in the study the outlier data were omitted.
The outlier data in this case cannot be simply omitted, because there was suspicion
of which a respondent has a blood glucose level that is extremely higher than the
normal blood glucose level.
The quantile regression method was applied in this case, since the method
is robust to outliers and does not need a normal distribution assumption. However,
the calibration modeling requires a variable reduction in advance because in general,
the calibration data is highly multi collinear in explanatory variables. The method
that can be used to solve this problem is the principal component analysis.
This study aimed to build a calibration model for predicting non-invasive
blood glucose level by using the principal component of Quantile Regression
method. The response variable (Y) was the result of invasive measurements in the
form of blood glucose levels (mg/dL). The explanatory variable (X) was the result
of non – invasive measurement in the form of a spectrum of residual light intensity
toward the time domain. This study applied 2017 and 2019 data with different tool
designs. In 2017, the design of the tool with the light captured by the sensor was
the light transmitted/passed by the bloods. In 2019, the tool designed sensor which
captured the light reflected by the bloods. Prior to modeling, data pre-processing
was carried out by two approaches. First, the approach was to summarize the area
under the residual intensity curve for each periode time (period area). Second, the
approach was to summarize the area under the residual intensity curve for each the
lamps were on (peak area). The model was constructed by various quantiles and
several numbers of principal components which were selected based on various
proportions of the cumulative variance.
Based on our analysis, overall RMSEP value from summarizing period area
data were steadier compared to summarizing of peak area. The sixth quantile had
RMSEP value of 0,0184, which was the best in predicting blood glucose level tested
non-invasively according to proportion of the cumulative variance of 90%.
However, our study showed weakness in prediction value which did not follow
actual data pattern. Moreover, the correlation between actual data and its prediction
still not strong enough Diabetes Melitus (DM), merupakan penyakit gangguan metabolisme menahun akibat pankreas tidak memproduksi cukup insulin (hormon yang mengatur gula darah) atau tubuh tidak dapat menggunakan insulin yang diproduksi secara efektif. Akibatnya, terjadi peningkatan konsentrasi glukosa di dalam darah (hiperglikemia). Jika DM tidak segera diatasi, maka dapat mengakibatkan komplikasi yang berujung pada kematian. DM merupakan salah satu masalah kesehatan global dengan keadaan darurat yang paling cepat berkembang pada abad ke-21, oleh karena itu perlu adanya tindakan pencegahan. Salah satunya dengan pengecekan kadar gula darah secara rutin. Pengecekan kadar gula darah biasanya dilakukan secara invasif (jarum suntik dan glukometer) yang bersifat melukai tubuh, membutuhkan waktu yang cukup lama, dan biaya yang cukup mahal. Permasalahan ini membuat tim biomarking IPB mencoba mengembangkan alat pemantau gula darah secara non-invasif yang bersifat tanpa melukai tubuh. Pengembangan alat pemantau ini memerlukan analisis model kalibrasi. Penelitian sebelumnya menunjukkan bahwa terdapat data pencilan pada hasil pengukuran kadar gula darah invasif. Dalam penelitian tersebut data pencilan dihilangkan. Data pencilan pada kasus ini tidak dapat dihilangkan begitu saja. Hal ini dikarenakan data pencilan dapat memberikan pengaruh yang berbeda-beda terhadap hasil keputusan. Metode regresi kuantil diterapkan pada kasus ini, mengingat bahwa metode tersebut kekar terhadap pencilan dan tidak membutuhkan asumsi distribusi normal. Pemodelan kalibrasi memerlukan pereduksian peubah terlebih dahulu, sebab secara umum data kalibrasi memiliki multikolinearitas yang tinggi antar peubah penjelas. Pada penelitian ini, metode yang digunakan untuk mengatasi masalah tersebut yaitu Analisis Komponen Utama. Penelitian ini bertujuan membangun model kalibrasi untuk memprediksi kadar gula darah non-invasif dengan metode Regresi Kuantil Komponen Utama. ... dst...