A carregar...
Publicação
Aplicação do modelo de Regressão Multinomial e algoritmos de aprendizagem automática na classificação de origem da espécie Pollicipes pollicipes
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 3.01 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
O rastreamento de produtos alimentares provenientes do mar é, atualmente, considerado
um problema de extrema importância e necessidade, tanto para os consumidores
como para os comerciantes, dada a globalização do mercado e os alertas recorrentes sobre
as questões de segurança alimentar. Este problema tem sido muito debatido, visto que
é comum e propício ocorrerem crimes de fraude, como a apanha ilegal. Por forma a
combater este problema, é possível desenvolver modelos capazes de classificar a origem
dos produtos com alto nível de precisão.
Neste estudo considerou-se a espécie Pollicipes pollicipes, vulgo percebes, muito abundante
nos habitats rochosos nas costas de Portugal continental e a norte de Espanha.
Dada a sua abundância e elevado valor económico associado, trata-se de um recurso com
grande interesse na indústria, ficando suscetível à apanha ilegal. Assim sendo, pretende-se
prever a origem geográfica dos percebes com base na sua composição geo- e bioquímica.
O conjunto de dados em análise é constituído por 176 observações de 6 origens distintas
(Corunha, Espanha; Berlengas, Estelas, Peniche Sul, Peniche Norte e Sines, Portugal),
contendo informação acerca de 46 elementos geo- e bioquímicos.
Para prever a origem dos percebes, recorreu-se a métodos de classificação como
Regressão Logística Multinomial, Árvores de Classificação e Random Forest. Ao estudar
a convergência dos modelos de previsão, com vista a avaliar que características dos
percebes são determinantes para a previsão da sua origem, foi também determinada a
importância associada a cada um dos preditores para cada um dos modelos ajustados.
Por fim, procedeu-se à validação dos modelos e posterior comparação dos métodos de
classificação.
Given the market globalization and the recurring alerts regarding food safety issues, seafoodtracing is an ongoing issue of extreme importance andnecessity forbothconsumers and traders. This problem has been much debated since the growing occurrences of fraud crimes, such as illegal harvesting. To combat this problem, it is possible to develop models capable of classifying the origin of products with a high level of precision. In this study,we relied on the species Pollicipes pollicipes, commonly knownas barnacles, very abundant in rocky habitats on the coasts of mainland Portugal and northern Spain. Given its abundance and high associated economic value, it is a resource of great interest to the industry and consequently susceptible to illegal harvesting. Therefore, we intend to predict the species’s geographical origin through its geochemical and biochemical characteristics. The dataset consists of 176 observations of barnacles from 6 different locations (Coruña, Spain; Berlengas, Estelas, Peniche Sul, Peniche Norte, and Sines, Portugal) containing information about 46 geo- and biochemical elements. Classification methods such as Multinomial Logistic Regression, Classification and Regression Trees, and Random Forest were used to predict the origin of barnacles. When studying the convergence of prediction models, to assess which barnacle characteristics are crucial for predicting their origin, the importance associated with each of the predictors for each of the adjusted models was also determined. Finally, the models were validated, and the classification methods were later compared.
Given the market globalization and the recurring alerts regarding food safety issues, seafoodtracing is an ongoing issue of extreme importance andnecessity forbothconsumers and traders. This problem has been much debated since the growing occurrences of fraud crimes, such as illegal harvesting. To combat this problem, it is possible to develop models capable of classifying the origin of products with a high level of precision. In this study,we relied on the species Pollicipes pollicipes, commonly knownas barnacles, very abundant in rocky habitats on the coasts of mainland Portugal and northern Spain. Given its abundance and high associated economic value, it is a resource of great interest to the industry and consequently susceptible to illegal harvesting. Therefore, we intend to predict the species’s geographical origin through its geochemical and biochemical characteristics. The dataset consists of 176 observations of barnacles from 6 different locations (Coruña, Spain; Berlengas, Estelas, Peniche Sul, Peniche Norte, and Sines, Portugal) containing information about 46 geo- and biochemical elements. Classification methods such as Multinomial Logistic Regression, Classification and Regression Trees, and Random Forest were used to predict the origin of barnacles. When studying the convergence of prediction models, to assess which barnacle characteristics are crucial for predicting their origin, the importance associated with each of the predictors for each of the adjusted models was also determined. Finally, the models were validated, and the classification methods were later compared.
Descrição
Palavras-chave
Regressão Logística Multinomial Árvores de Classificação Random Forest Classificação