Application of Ensemble Machine Learning Methods to the Classification and Interpretation of Raman Spectra of Adipose Tissue During Enzymatic Hydrolysis by Lipase

Abstract

Raman spectroscopy (RS) in combination with machine learning methods was used to solve the problem of the classification of adipose tissue before and after exposure to lipase. Differences between the classes were identified using principal component analysis. Hyperparameters such as the number of trees (n_estimators) and the maximum depth (max_depth) of the random forest (RF) and gradient boosting (GB) ensemble models are optimized for balanced accuracy and training time. This made ensured obtaining models with high predictive ability. In this method, models with 50 decision trees with a maximum depth of 3 were trained. The weighted accuracies of the RF and GB models were 96.4 ± 8.7% and 92.9 ± 11.3%, respectively. While the final prediction of the RF model was affected by intensities at 37 wavenumbers, including those corresponding to vibrations of the C=C bonds of unsaturated fatty acids, as well as of aliphatic C−C bonds and the carbonyl group >C=O, the prediction of the GB was based predominantly on the intensity at 1650 cm⁻¹ (C=C, importance 95.4%). The results obtained are consistent with the formation of free fatty acids, mono- and diglycerides as a result of the enzymatic hydrolysis of adipose tissue.