Aprendizado de Máquina Aplicado à Espectroscopia: Uma Abordagem Não Invasiva para Sexagem de Aves

Abstract

Identifying the sex of birds is of utmost importance for marketing, veterinarians, and breeders. DNA (deoxyribonucleic acid) based methodologies have been developed. However, although it provides highly sensitive identification, it is a time-consuming method that often requires specific laboratories. In this regard, some studies present spectroscopy techniques associated with machine learning for more accurate analysis. This study presents analyses of spectroscopic measurements by Fourier Transform Infrared Spectroscopy (FTIR) associated with machine learning algorithms as a non-invasive technique for determining bird sex. The spectra analyzed were from the species Oryzoborus maximiliani (Bicudo) from 4000 to 600 cm^(-1). The methodology involved collecting spectra from bird feathers, processing the data using SNV (Standard Normal Variate), and submitting them to Principal Component Analysis (PCA) to reduce and identify significant variables to be passed through the unsupervised learning algorithm, with K-means being used to select the best components. The selection of PC combined with the leave-one-out cross-validation (LOOCV) method for the LDA (Linear Discriminant Analysis), SVM (Support Vector Machine), and KNN (K-Nearest Neighbors) classifiers showed significant and promising results, with SVM standing out, achieving an accuracy of 99.17%. This result demonstrated the efficiency of the applied methodology.