CORAL: machine learning e matrizes de rotação aplicados a resolução de sistemas multivariados

• Autor(a) principal: Torquato, Igor Ferreira
• Data de Publicação: 2022
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/20.500.14289/16054
• topic: Combinação linear; Ambiguidade rotacional; CORAL; MCR-ALS; PCA; Jupyter; CIENCIAS EXATAS E DA TERRA::FISICA

CORAL (Curve ResOlution for dAta anaLysis) is a Python-based library of chemometric tools for multivariate spectral decomposition of large datasets, especially those where the techniques involved obey Beer's law or a form of linear data combination. . Researchers can use it in the Jupyter environment to address challenges related to the large number of spectra generated by rapid data acquisition during time-resolved catalytic reaction studies. In addition to allowing the establishment of a unique experimental notebook for controlling the beamlines of the fourth generation particle accelerator, Sirius, and analyzing data in Jupyter. Despite being versatile, multivariate techniques, such as MCR-ALS (Multivariate Curve Resolution with Alternating Least Squares) and PCA (Principal Component Analysis), present problems inherent to spectral decomposition, since in each decomposition there are multiple responses. that satisfy the system of matrix equations the variability of responses is known as Rotational Ambiguity (ARs). In this aspect, the study of ARs can reveal important information about the studied system, such as the error associated with each pure compound found by the multivariate methods. The present project intends to present the developments of CORAL, complement the studies of ARs and evaluate how the constraints (constraints) affect the spectral decomposition responses, applying the study of plausible solution areas and transformation matrices to observe the evolution of solution spaces for a dataset with two components. Thus, the objective of this project is to add a new tool to CORAL so that its users can enjoy a more complete analysis of the data sets, in order to guarantee the choice of the best solutions. Furthermore, preliminary analyzes suggest that the method can be used as an estimate of the error associated with rotational ambiguity and the spectra and concentrations can be an initial estimate of the MCR-ALS.