Publicação
Evaluating the role of environmental variables on shellfish biotoxin contamination via supervised learning
| Resumo: | The production and harvest of shellfish is threatened by harmful algal bloom events that can contaminate these filter-feeding organisms with marine biotoxins. Several studies have been carried out on this topic, but harmful algal blooms are a complex phenomena that still require further investigation to better understand its occurrence and its impact on shellfish contamination. After studies in the Portuguese mainland coast regarding shellfish contamination by marine biotoxins and coastal upwelling recognition through remotely sensed sea surface temperature images, this dissertation aims at broadening the knowledge on this area by studying shellfish contamination in several shellfish production regions in the Portuguese coast and assessing the role on this phenomenon of several environmental drivers including meteorological, hydrodynamic, water properties and coastal upwelling variables. Combining data acquired from previous works and partner institutions, this dissertation focuses on developing an appropriate experimental protocol capable of constructing several machine learning models capable of predicting shellfish contamination, exploring different approaches and algorithms. The work developed included an initial data preprocessing and analysis stage, that merged the data from distinct spatio-temporal sources and selected the best regions and variables. The models for shellfish contamination prediction were developed considering both classification and regression approaches, exploring the predictions as contamination classes or as biotoxin concentration levels. The algorithms used in this work, Random Forest and Support Vector Machine, were selected due to adequacy of their characteristics to the problem and past uses in the literature. The classification approach proved the most successful, correctly predicting most shellfish contamination data cases across the different zones. The inclusion of environmental variables in various combinations proved beneficial for certain models and regions. |
|---|---|
| Autores principais: | Oliveira, Manuel Bernardo Ribeiro de |
| Assunto: | Shellfish Contamination Environmental Drivers Random Forest Support Vector Machine |
| Ano: | 2023 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | The production and harvest of shellfish is threatened by harmful algal bloom events that can contaminate these filter-feeding organisms with marine biotoxins. Several studies have been carried out on this topic, but harmful algal blooms are a complex phenomena that still require further investigation to better understand its occurrence and its impact on shellfish contamination. After studies in the Portuguese mainland coast regarding shellfish contamination by marine biotoxins and coastal upwelling recognition through remotely sensed sea surface temperature images, this dissertation aims at broadening the knowledge on this area by studying shellfish contamination in several shellfish production regions in the Portuguese coast and assessing the role on this phenomenon of several environmental drivers including meteorological, hydrodynamic, water properties and coastal upwelling variables. Combining data acquired from previous works and partner institutions, this dissertation focuses on developing an appropriate experimental protocol capable of constructing several machine learning models capable of predicting shellfish contamination, exploring different approaches and algorithms. The work developed included an initial data preprocessing and analysis stage, that merged the data from distinct spatio-temporal sources and selected the best regions and variables. The models for shellfish contamination prediction were developed considering both classification and regression approaches, exploring the predictions as contamination classes or as biotoxin concentration levels. The algorithms used in this work, Random Forest and Support Vector Machine, were selected due to adequacy of their characteristics to the problem and past uses in the literature. The classification approach proved the most successful, correctly predicting most shellfish contamination data cases across the different zones. The inclusion of environmental variables in various combinations proved beneficial for certain models and regions. |
|---|