A novel robust PLS regression method inspired from boosting principles: RoBoost-PLSR.

Area of Science:

• Chemometrics
• Statistical Modeling
• Machine Learning

Background:

• Partial Least Square regression (PLSR) models are susceptible to performance degradation caused by outlying samples.
• Outliers can lead to unstable PLSR models and reduced predictive accuracy, necessitating robust calibration methods.

Purpose of the Study:

• To introduce a novel robust Partial Least Square regression algorithm, RoBoost-PLSR, designed to mitigate the impact of outlying samples.
• To evaluate the performance of RoBoost-PLSR against standard PLSR and a reference robust method (PRM).

Main Methods:

• The proposed RoBoost-PLSR algorithm utilizes a series of one latent variable weighted PLSR steps, inspired by boosting principles.
• The method involves downweighting outlier samples during the calibration process based on an estimated inconsistency measurement.

Main Results:

• RoBoost-PLSR demonstrated resilience against tested outliers across simulated and real datasets.
• The algorithm achieved performance levels comparable to standard PLSR models calibrated on data free from outliers.

Conclusions:

• RoBoost-PLSR offers an effective solution for robust calibration of PLSR models in the presence of outliers.
• The proposed method maintains high predictive performance, even with contaminated datasets, making it a valuable tool in chemometrics.