PLS Predictive Model for In-Vivo Non-Invasive Finger Touch Blood Glucose NIR Spectrosensor

Diabetes is one of the non-communicable disease that has high prevalence trends. The use of finger pricking to monitor the glucose level is painful thus non-invasive alternative is needed. Non-invasive finger touch blood glucose NIR spectrosensor was utilized for the in-vivo glucose monitoring on patients at Hospital Universiti Kebangsaan Malaysia (HUKM). The development of predictive model for the glucose monitoring was based on partial least square (PLS) algorithm. Different type of preprocesses were selected to check the performance of the model with different method of signal preprocessing. The optimum number of variable when performing variable selection is 744. The R
²
C and R
²
p acquired were 0.2492 and 0.1734 respectively. The RMSEC and RMSEP for the model were 3.0324 and 2.9901 for the combination of preprocessing of generalized least square (Gls) weighting and autoscale (As). Interval PLS (iPLS) was implemented for the wavelength extraction to enhance the predictive model. R
²
p for 500 variables (wavelength points) shown the best result with a value of 0.2390. RMSEP also has decreased by to 2.8221 from the previous model 2.9956.