An adaptive EEMD-machine learning algorithm for multi-matrix drug concentration prediction using miniature mass spectrometry

Abstract

In this study, an innovative approach combining Machine Learning (ML) with Ensemble Empirical Mode Decomposition (EEMD) was proposed to predict lamotrigine concentrations in actual samples, improving detection performance in complex matrices. EEMD decomposed the mass spectrometry data to extract Intrinsic Mode Functions (IMFs), enabling separation of noise from key signal features. Ridge Regression (RR) addressed multicollinearity among high-dimensional IMF features and enhanced model generalization via L2 regularization. ML was further applied to optimize the key EEMD parameter (ensemble number K),thereby improving both decomposition quality and prediction accuracy. Experimental validation showed that the method achieved over 90 % prediction accuracy in three types of blind samples (PBS, rabbit blood, and human matrix), with improved Relative Standard Deviation (RSD). These results confirm the method’s precision and robustness in diverse biological matrices. Compared to traditional techniques, the proposed approach delivers marked improvements in both accuracy and stability, can supporting more reliable drug concentration monitoring for clinical applications.