Efficient Decision Tree Based Data Selection and Support Vector Machine Classification

Detection of Advanced Glycosylation End Products in the Cornea Based on Molecular Fluorescence and Machine Learning

Advanced glycosylation end products (AGEs) are continuously produced and accumulated in the bodies of diabetic patients. To effectively predict disease trends in diabetic patients, a corneal fluorescence detection device was designed based on the autofluorescence properties of AGEs, and corneal fluorescence measurements were performed on 83 volunteers. Multiple linear regression (MLR), extreme gradient boosting (XGBoost), support vector regression (SVR), and back-propagation neural network (BPNN) were used to predict the human AGE content. Physiological parameters which may affect corneal AGE content were collected for a correlation analysis to select the features that had a strong correlation with the corneal concentration of AGEs to participate in modeling. By comparing the predictive effects of the four models in the two cases of a single-input feature and a multi-input feature, it was found that the model with the single-input feature had a better predictive effect. In this case, corneal AGE content was predicted by a single-input SVR model, with the average error rate (AER), mean square error (MSE), and determination coefficient R-squared (R²) of the SVR model calculated as 2.43%, 0.026, and 0.932, respectively. These results proved the potential of our method and device for noninvasive detection of the concentration of AGEs in the cornea.