Fusion of EEMD with extra trees-gradient boosting feature extraction for quantification analysis of polysaccharides...

Near-infrared spectroscopy enables rapid quantitative prediction of plant constituents; however, extracting informative features from complex and overlapping spectral data remains a persistent challenge. To address this issue, this study proposes a framework that integrates Ensemble Empirical Mode Decomposition (EEMD) with an Extra Trees-Gradient Boosting Feature Extraction (EGFE) method. A total of 90 turnip samples (70 for calibration, 20 for validation) were analyzed to predict polysaccharides and saponins contents. The spectral data were decomposed via EEMD into multiple intrinsic mode functions (IMFs) and a residual component. Feature importance was then evaluated using a composite scoring strategy, and the optimal subset was selected by sequential forward selection (SFS). The developed EEMD-EGFE-PLSR model achieved coefficients of determination (R²) of 0.7734 for polysaccharides and 0.9027 for saponins. These results indicate that the proposed framework provides an effective approach for the quantitative determination of bioactive compounds in Xinjiang turnip.