## Abstract

Near-Infrared (NIR) spectroscopic method is applied to analyze the glucose concentrations in aqueous and whole blood matrices. The major absorption bands influenced by the changes in glucose concentration were observed. These absorption spectra were analyzed by partial least squares (PLS) regression to compress the number of spectral components and establish a regression model. Cross-validation method is used to evaluate the performance of the PLS regression algorithm and determine the number of PLS regression factors in selected spectral range. To facilitate the searching of the optimal control variables, the spectral range and the number of PLS regression factors; a control variable search method is conducted by using regression factor map (RFM) and error surface map (ESM). The number of PLS regression factors and the mean-percentage error of cross-validation (MPECV) can be identified and verified in the RFM and ESM for the spectral range used for the PLS regression procedure. The optimal control variables in NIR spectral range for aqueous matrix: from 0.915 to 1.335 mm, PLS regression factor number 9; for whole blood matrix: from 1.105 to 1.335 mm, PLS regression factor number 10. The MPECV for glucose concentration in aqueous and whole blood matrices was 7.36 % and 9.20 %, respectively.

Original language | English |
---|---|

Pages (from-to) | 195-204 |

Number of pages | 10 |

Journal | Biomedical Engineering - Applications, Basis and Communications |

Volume | 12 |

Issue number | 4 |

State | Published - 25 Aug 2000 |

## Keywords

- Cross-validation
- NIR Spectroscopy
- PLS