Canine visceral leishmaniasis diagnosis by UV spectroscopy of blood serum and machine learning algorithms

Multi-resistant diarrheagenicEscherichia coliidentified by FTIR and machine learning: a feasible strategy to improve the group classification

“…Here, the main PCs were determined by the overall accuracy achieved in a Leave One Out Cross-Validation (LOOCV) test using Linear Discriminant Analysis (LDA). 60 The main PCs found for our data were: PC1 and PC2 for the 4000 to 800 cm −1 range, responsible for 62.83% of data variance; 49 rst PCs for the 3000 to 2800 cm −1 range, responsible for 99.9% of data variance; and PC3, PC4, PC9, PC18, PC23, and PC36 for the 1800 to 800 cm −1 range responsible for 11.30% of data variance.…”

“…We found PC1, PC2, and PC3 responsible for 79.3%, 89.7%, and 81.7% of data variance at 4000 to 800 cm −1 , 3000 to 2800 cm −1 , and 1800 to 800 cm −1 ranges, respectively; besides this great contribution for data variance, previous studies have shown that the rst PCs can be ignored, and high order PCs can be used to improve the group classication for ML algorithms. 60 Usually, the proper choice of spectral range helps to improve group classication and clustering formation 61 since we use only spectral information that most contributes to clustering instead of those with highly correlated data, which hinders cluster formation. But here, we couldn't succeed with this strategy probably because of the high similarities between the groups involved and the highly correlated data present.…”

“…However, low data variance PCs can still be helpful for group classication and better inuence the algorithm performance, as demonstrated in the literature. 60 The RFE-PCs were submitted to machine learning algorithms (DA, SVM, and KNN) to build a prediction model for sample classication; the overall accuracy achieved for each model by using different functions for classication in the LOOCV test is summarized in Fig. 6(a).…”

“…We found PC1, PC2, and PC3 responsible for 79.3%, 89.7%, and 81.7% of data variance at 4000 to 800 cm −1 , 3000 to 2800 cm −1 , and 1800 to 800 cm −1 ranges, respectively; besides this great contribution for data variance, previous studies have shown that the first PCs can be ignored, and high order PCs can be used to improve the group classification for ML algorithms. 60 …”

“…However, low data variance PCs can still be helpful for group classification and better influence the algorithm performance, as demonstrated in the literature. 60 …”