Related Experiment Videos
Elimination of the uninformative calibration sample subset in the modified UVE(Uninformative Variable
J Koshoubu1, T Iwata, S Minami
1JASCO Technical Research Laboratory Corporation, Hachioji, Tokyo, Japan.
Summary
A new algorithm improves Partial Least Squares (PLS) models by removing uninformative wavelength variables and samples. This enhances predictive accuracy in spectral data analysis.
Area of Science:
- Chemometrics
- Spectroscopy
- Data Analysis
Background:
- Partial Least Squares (PLS) models are widely used for predictive modeling.
- Improving the predictive ability of PLS models is crucial for accurate analysis.
- Identifying and removing uninformative data can enhance model performance.
Purpose of the Study:
- To develop a novel algorithm for increasing the predictive ability of PLS models.
- To eliminate uninformative samples and wavelength variables from calibration datasets.
- To enhance the precision of PLS models through data refinement.
Main Methods:
- Developed a new algorithm integrating modified Uninformative Variable Elimination-PLS (UVE-PLS).
- Eliminated uninformative wavelength variables in the first stage.
- Removed uninformative samples based on prediction error exceeding 3-sigma using leave-one-out cross-validation.
Main Results:
- The algorithm successfully identified and removed uninformative wavelength variables.
- Uninformative samples were effectively eliminated based on prediction error thresholds.
- The refined PLS model demonstrated improved predictive capabilities.
Conclusions:
- The proposed algorithm enhances PLS model predictive ability by removing uninformative variables and samples.
- This method offers a precise approach to constructing robust PLS models.
- The algorithm's utility was validated using mid-infrared spectral datasets.