Influence of PCA components on glucose prediction using non-invasive technique

A truly noninvasive blood glucose-sensing device could revolutionalize diabetes treatment and coupled with advances in microelectronics can improve compliance with recommended glucose levels and greatly influence the cost of diabetes monitoring. The paper describes a Multivariate PLSR system for blood glucose prediction by considering 5 major variants in human blood i.e Glucose, Alanine, Ascorbate, Lactate and Urea. The typical biological system that resembles human blood tissue consisting of five major constituents has been tested on the Partial Least Square Regression (PLSR) model reported by the authors elsewhere. Multivariate PLSR model is experimentally validated for 12 templates recorded using Schimatzu FTIR 8400S in the range 400cm⁻¹ to 5000 cm⁻¹. The model was validated using 2 approaches, namely, Root Mean Square Error (RMSE) and Clark Error Grid Analysis (CEGA) plot. The dependence of variate prediction on Principal Component Analysis(PCA) factors is also described and analyzed in detail.