[Rapid and robust partial least squares regression and its application to NIR spectroscopy analysis]

Area of Science:

• Analytical Chemistry
• Chemometrics

Context:

• Near-infrared spectroscopy (NIRS) is a key indirect analytical technique for quantitative analysis.
• Traditional NIRS methods, like SIMPLS, struggle with low sensitivity and outlier data.
• Robust modeling is crucial for accurate analysis of complex, high-dimensional datasets.

Purpose:

• To develop a robust SIMPLS algorithm for NIRS quantitative analysis.
• To address the limitations of traditional SIMPLS, particularly its sensitivity to outliers.
• To improve the accuracy and reliability of NIRS models in the presence of abnormal data.

Summary:

• A novel kurtosis-SIMPLS method is proposed, integrating robust covariance estimation and outlier detection.
• This method utilizes projections maximizing/minimizing kurtosis to identify and mitigate outlier impacts.
• The kurtosis-SIMPLS approach was applied to NIRS analysis, demonstrating superior performance over standard SIMPLS.

Impact:

• The kurtosis-SIMPLS method effectively detects outliers with reduced computational cost.
• It offers enhanced prediction performance and stable results for normal samples in NIRS.
• This advancement leads to more reliable quantitative analysis using NIRS techniques.