Integrating Biomarkers from Correlation Analysis and Metric Learning for Hair Loss Classification

This study investigates the correlation patterns of eight biomarkers in 35 patients with hair loss and 35 healthy controls, identifying significant relationships such as a strong positive correlation between C-reactive Protein and General Urine (r=0.858, p<0.01) and a negative correlation between Ferritin and Zinc (r=-0.353, p=0.032). The study also evaluates classification methods, finding that the LMNN algorithm, which learns an optimal Mahalanobis distance, achieves a 95% accuracy in distinguishing between hair loss patients and controls, outperforming the standard k-Nearest Neighbors method. This suggests that integrating correlation analysis with metric learning can enhance the detection and understanding of alopecia.