A carregar...
Publicação
MÉTODOS DE APRENDIZAGEM AUTOMÁTICA PARA PREDIÇÃO DE EVENTOS DE COVID-19
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 8.85 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
O aparecimento da COVID-19, uma doença viral e infeciosa que apresenta uma alta
taxa de propagação, resultou em grandes pressões sobre os sistemas de saúde, esgotando
médicos, recursos e provocando a exaustão dos profissionais de saúde da linha da frente.
Devido às suas consequências devastadores e ao seu desconhecimento, começaram-se a
recolher e analisar dados dos pacientes que tinham testado positivo à COVID-19, tanto
dos sintomas, como das suas COVID-19, variante do vírus, dados demográficos, dados
geográficos do local de infeção, assim como as datas de infeção e de início de sintomas,
para se tentar ajudar as unidades hospitalares a selecionarem os pacientes que deveriam
ter um atendimento prioritário. Uma vez que estes conjuntos de dados rapidamente
se tornaram gigantescos, utilizaram-se variados métodos, também de Aprendizagem
Automática. Estes métodos, têm como principal objetivo a perceção de padrões e o
relacionamento de dados, de uma forma automática a partir dos mesmos. Os métodos de
Aprendizagem Automática, fazem uso de métodos estatísticos e probabilísticos de forma
a criar algoritmos que consigam aprender diretamente de uma parte de dados, que são
usados para treinar e avaliar os modelos conforme o seu desempenho preditivo. Estes
métodos têm tido uma ampla utilização para problemas de classificação e em especial
foram muito empregues no contexto da COVID-19, uma vez que desta doença nova não
se sabia praticamente nada sobre a sua evolução ao longo do tempo, e aplicando-se
diversos modelos de AprendizagemAutomática aos dados de saúde recolhidos, tentou-se
fazer previsões para o número de casos diários, assim como também para a identificação
dos pacientes, consoante os sintomas e as comorbilidades que estes apresentavam, que
fossem de atendimento mais premente nos hospitais. Assim, o presente trabalho tem como
objetivo a aplicação de diversos métodos de Aprendizagem Automática, entre os quais
Regressão Logística, Modelos Aditivos Generalizados, Árvores de Classificação, Florestas
Aleatórias e Redes Neuronais, em dados Portugueses de saúde de COVID-19, de forma a
avaliar cada um destes modelos em termos preditivos e também tem como objetivo saber
quais os sintomas e as doenças que estão mais relacionadas com a morte por COVID-19.
The emergence of COVID-19, a viral and infectious disease that has a high rate of spread, has resulted in great pressures on health systems, depleting medical doctors, resources and causing the exhaustion of frontline health professionals. Due to its devastating consequences and lack of knowledge about it, data began to be collected and analyzed from patients who had tested positive for COVID-19, both on symptom data, comorbidities data, virus variant, demographic data, geographic data the site of infection, as well as the dates of infection and onset of symptoms, to try to help hospital units to select patients who should receive priority care. Once these datasets quickly became huge, they begam to be analysed with several methods, including various Machine Learning methods. These methods have as main objective the revelation of patterns and data relationships, automatically. Machine learning methods make use of statistics and probability in order to create the learning methods that are directly used to train and evaluate models according to their predictive performance. Machine Learning methods have a wide use for classification and especially have been much used in COVID-19 context, since this disease was new and practically nothing was known about its evolution in time. Several Machine Learning models were applied to the collected health data, in order to try to make predictions for the number of daily cases, as well as for the identification of most potentially problematic patients, depending on their symptoms and comorbilities, so that hospitals would pay more attention to those. Thiswork have the objective of applying several Machine Learning methods, including Logistic Regression, Generalized Additive Models, Group LASSO, Classification Trees, Random Forests and Neural Networks, to the Portuguese COVID-19 data, in order to evaluate each one of these models in predictive terms and also have the objective to know which symptoms and comorbidities are most related with mortality from COVID-19.
The emergence of COVID-19, a viral and infectious disease that has a high rate of spread, has resulted in great pressures on health systems, depleting medical doctors, resources and causing the exhaustion of frontline health professionals. Due to its devastating consequences and lack of knowledge about it, data began to be collected and analyzed from patients who had tested positive for COVID-19, both on symptom data, comorbidities data, virus variant, demographic data, geographic data the site of infection, as well as the dates of infection and onset of symptoms, to try to help hospital units to select patients who should receive priority care. Once these datasets quickly became huge, they begam to be analysed with several methods, including various Machine Learning methods. These methods have as main objective the revelation of patterns and data relationships, automatically. Machine learning methods make use of statistics and probability in order to create the learning methods that are directly used to train and evaluate models according to their predictive performance. Machine Learning methods have a wide use for classification and especially have been much used in COVID-19 context, since this disease was new and practically nothing was known about its evolution in time. Several Machine Learning models were applied to the collected health data, in order to try to make predictions for the number of daily cases, as well as for the identification of most potentially problematic patients, depending on their symptoms and comorbilities, so that hospitals would pay more attention to those. Thiswork have the objective of applying several Machine Learning methods, including Logistic Regression, Generalized Additive Models, Group LASSO, Classification Trees, Random Forests and Neural Networks, to the Portuguese COVID-19 data, in order to evaluate each one of these models in predictive terms and also have the objective to know which symptoms and comorbidities are most related with mortality from COVID-19.
Descrição
Palavras-chave
COVID-19 Aprendizagem Automática Regressão Logística Modelo Aditivo Generalizado LASSO de Grupo Árvores de Classificação