A comparative study between PCR and PLSR in a tapered optical fiber sensor for acetone detection

The use of multivariate analysis techniques and prediction models represents a novel alternative in sensors based on optical fibers, since they have a great potential to estimate the properties and quality of the response of optical fiber sensors. This work focuses on comparing and determining the accuracy of prediction models such as multivariate projection to latent structures regression (PLSR) and principal component regression (PCR) techniques applied to tapered optical fiber sensors as well as experimental studies of the different sensing films used, such as polydimethylsiloxane (PDMS), polymethyl methacrylate (PMMA), Apiezon T (ApT) and Apiezon L (ApL) for acetone detection. Acetone is a biomarker of diabetes mellitus which is found in concentrations in the order of 1.8 ppm in the human breath of diabetic patients. The results showed that the sensor developed with the PMMA sensor film improved the limit of detection (LOD) up to 5.56 ppm using PLSR with four latent variables compared to PCR using the same number of components.