Partial least squares and random sample consensus in outlier detection.

Area of Science:

• Chemometrics
• Data Mining
• Statistical Modeling

Background:

• Partial Least Squares (PLS) is a widely used regression method.
• Outlier detection is crucial for robust data analysis and model building.
• Existing methods like leave-one-out cross-validation can be computationally intensive.

Purpose of the Study:

• To introduce a novel outlier detection method for PLS models.
• To enhance the robustness and efficiency of outlier identification in chemometric datasets.
• To compare the proposed method against established techniques.

Main Methods:

• Developing a novel outlier detection algorithm based on random sample consensus (RSC) for PLS.
• Generating multiple PLS solutions from random subsamples.
• Assessing the consistency of subsample solutions against the complete dataset.

Main Results:

• The proposed RSC-based PLS outlier detection method demonstrates high efficiency.
• Comparative studies on simulated and real-world near-infrared (NIR) pharmaceutical data show superior performance.
• The method outperforms standard PLS, RSIMPLS, and PRM in outlier detection accuracy and consistency.

Conclusions:

• The novel RSC-based PLS outlier detection method offers a significant advancement.
• This technique provides a robust and efficient solution for identifying outliers in complex datasets.
• The findings suggest broad applicability in chemometrics and related fields.