Performing sequential forward selection and variational autoencoder techniques in soil classification based on laser-induced breakdown spectroscopy

The feasibility and accuracy of several combination classification models, , quadratic discriminant analysis (QDA), random forest (RF), Bernoulli naïve Bayes (BNB), and support vector machine (SVM) classification models combined with either sequential feature selection (SFS) or dimensionality reduct...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Analytical methods Ročník 13; číslo 41; s. 4926
Hlavní autoři: Harefa, Edward, Zhou, Weidong
Médium: Journal Article
Jazyk:angličtina
Vydáno: England 28.10.2021
Témata:
Bayes Theorem
Lasers
Soil
Spectrum Analysis
Support Vector Machine
ISSN:1759-9679, 1759-9679
On-line přístup:Zjistit podrobnosti o přístupu
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:The feasibility and accuracy of several combination classification models, , quadratic discriminant analysis (QDA), random forest (RF), Bernoulli naïve Bayes (BNB), and support vector machine (SVM) classification models combined with either sequential feature selection (SFS) or dimensionality reduction methods, for classifying soil with laser-induced breakdown spectroscopy (LIBS) had been explored in this study. Each algorithm combination was compared to assess their classification performance. After eliminating the irrelevant features of the data using sequential feature selection (SFS), the performances were all improved for the studied four classification models, and the best accuracy reached 97.88% by SFS-SVM. The dimensions of the data were then reduced using variational autoencoder (VAE), truncated singular value decomposition (TSVD), and isometric mapping (Isomap), respectively. The classification accuracy improved for all combination models with dimensionality reduction, and impressive accuracies of 98.12% from TSVD-SVM and 98.24% from VAE-SVM were obtained. These results demonstrate an effective way to reduce uncorrelated features, high dimensionality, and redundant information in the LIBS dataset. In addition, coupling classification models with feature selection and dimensionality reduction techniques could significantly optimize the classification performance of LIBS.
Bibliografie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 23
ISSN:1759-9679
1759-9679
DOI:10.1039/d1ay01257f