<![CDATA[General Background: Microcirculation disorders represent an early marker of chronic health conditions, yet existing detection approaches predominantly rely on invasive and resource-intensive procedures. Specific Background: Recent advances in wearable technology enable noninvasive microcirculation monitoring through Laser Doppler Flowmetry and Fluorescence Spectroscopy signals, which generate complex, nonstationary, and high-dimensional data that challenge conventional analytical methods. Knowledge Gap: Despite the proven capability of Light Gradient Boosting Machine models for wearable physiological data, limited studies have systematically combined feature selection, Bayesian hyperparameter optimization, and cohort-based validation for microcirculation condition detection using LDF-FS data. Aims: This study aims to optimize LightGBM performance for microcirculation condition detection by integrating feature importance–based selection and Bayesian hyperparameter tuning within a Stratified Group K-Fold validation framework. Results: Feature dimensionality was reduced from 34 to 22 informative variables, resulting in improved classification performance, with the optimized model achieving a ROC-AUC of 0.8632, accuracy of 88.04%, and recall of 80.00%. SHAP-based analysis identified age, body mass index, and skin temperature as dominant physiological predictors. Novelty: The study presents an integrated optimization pipeline combining feature selection, Bayesian optimization, and subject-level validation on wearable LDF-FS data. Implications: The findings support the potential of optimized LightGBM models as interpretable and reliable components of noninvasive wearable-based microcirculation monitoring systems. Highlights • Feature selection reduced dimensionality while maintaining robust classification performance• Bayesian optimization improved sensitivity in detecting microcirculation conditions• SHAP analysis revealed dominant demographic and physiological predictors Keywords Bayesian Optimization; Microcirculation Detection; Feature Selection; LightGBM; Wearable LDF-FS]]>
