Application of Multivariate Curve Resolution analysis and Multilayer Perceptron for Classification of Raman Spectra of the Skin

This study examines the possibility of using a multilayer perceptron, as well as the expediency of reducing the data dimension using multivariate curve resolution analysis to classify the Raman spectra of the skin. Classification models have been developed for three cases: healthy skin and skin with diseases, benign neoplasms and malignant neoplasms, and malignant melanoma and pigmented nevi. To study the dimensionality reduction of the data, the method of multivariate curve resolution using the alternating least squares (MCR-ALS) method was used. Data were obtained on eight, ten, fifteen and thirty skin components and their relative concentrations in the tested samples, which were used as classification parameters. The effectiveness of the classification models with a different number of parameters was compared. The ROC AUC of the classification varies from 0,60 to 0,87.