| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 6.26 MB | Adobe PDF | |||
| 1.74 KB | License |
Advisor(s)
Abstract(s)
No ano de 2019 iniciou-se uma pandemia, doença infeciosa que se espalha pela população numa grande escala geográfica, da pneumonia coronavirus 2019 (COVID-19), causada pelo SARS-Cov-2 (severe acute respiratory sindrome coronavirus 2), um auxiliar ao combate da crescente evolução deste vírus seria construir um sistema que agregue informação que está dispersa em várias fontes de informação diferentes e usar essa informação para prever essa evolução usando modelos de Machine Learning. Através de pesquisas verificou-se que nem todos os modelos são indicados para este tipo de previsão, sendo que modelos lineares como por exemplo a regressão linear não conseguem produzir previsões de qualidade, desta forma os melhores modelos são aqueles que foram projetados para previsões ao longo do tempo, ou transformar os dados de forma que estes possam ser utilizados para este tipo de previsões na maioria dos modelos. Através de pesquisa foi possível identificar modelos que poderiam obter bons resultados, que medidas utilizar para avaliar estes modelos, assim como variáveis que poderão ajudar os modelos a terem um desempenho melhor nas previsões. Após todo o trabalho de pesquisa o projeto procurou realizar vários testes nas previsões da evolução do covid-19 em Portugal, ou seja, nas previsões do número de casos confirmados, casos ativos, recuperados e óbitos. Os primeiros testes têm como objetivo identificar as caraterísticas que fornecem aos modelos um melhor desempenho para prever o dia N+1, sendo N o dia atual. Após obter as carateristicas que produzem melhores resultados para previsões N+1, utilizou-se os modelos e essas caraterísticas para executar previsões até N+5, em que se usa as próprias previsões para gerar previsões, desta forma verificou-se que os modelos que eram melhores nas previsões N+1, não são tão bons em previsões até N+5. Por fim conseguiu-se implementar um sistema intuitivo, totalmente construído em linguagem R, que permite ao utilizador aceder a várias funcionalidades, como por exemplo analisar os dados atuais da pandemia, analisar a evolução da pandemia desde o seu início até ao dia atual, analisar as previsões em Portugal e suas regiões, assim como comparar as previsões.
In 2019 a pandemic began, an infectious disease that spreads throughout the population on a large geographic scale of pneumonia coronavirus 2019 (COVID-19), caused by SARS-Cov-2 (severe acute respiratory syndrome coronavirus 2), an auxiliary to the combat of evolution of this virus would be to build a system that aggregates information that is dispersed across many different information sources and use that information to predict that evolution using Machine Learning models. Through research it was found that not all models are suitable for this type of prediction, and linear models such as linear regression cannot produce quality predictions, thus the best models are those that were designed for predictions along of time intervals, or transform the data so that they can be used for this type of prediction in most models. Through research, it was possible to identify models that could obtain good results, which measures to use to evaluate these models, as well as variables that could help the models to have a better performance in the predictions. After all the research work, the project tried to carry out several tests in the predictions of the evolution of covid-19 in Portugal, that is, in the predictions of the number of confirmed cases, recovered cases and deaths. The first tests aim to identify the characteristics that provide the models with a better performance to predict day N+1, with N being the current day. After obtaining the characteristics that produce the best results for N+1 predictions, we used the models and these characteristics to perform predictions up to N+5, in which the predictions themselves are used to generate predictions. were the best models at N+1 predictions, didn’t perform so good at predictions up to N+5. Finally, we managed to implement an intuitive system, entirely built in R language, which allows the user to access various features, such as analyzing current pandemic data, analyzing the evolution of the pandemic from its beginning to the present day, analyzing forecasts in Portugal and its regions, as well as comparing forecasts.
In 2019 a pandemic began, an infectious disease that spreads throughout the population on a large geographic scale of pneumonia coronavirus 2019 (COVID-19), caused by SARS-Cov-2 (severe acute respiratory syndrome coronavirus 2), an auxiliary to the combat of evolution of this virus would be to build a system that aggregates information that is dispersed across many different information sources and use that information to predict that evolution using Machine Learning models. Through research it was found that not all models are suitable for this type of prediction, and linear models such as linear regression cannot produce quality predictions, thus the best models are those that were designed for predictions along of time intervals, or transform the data so that they can be used for this type of prediction in most models. Through research, it was possible to identify models that could obtain good results, which measures to use to evaluate these models, as well as variables that could help the models to have a better performance in the predictions. After all the research work, the project tried to carry out several tests in the predictions of the evolution of covid-19 in Portugal, that is, in the predictions of the number of confirmed cases, recovered cases and deaths. The first tests aim to identify the characteristics that provide the models with a better performance to predict day N+1, with N being the current day. After obtaining the characteristics that produce the best results for N+1 predictions, we used the models and these characteristics to perform predictions up to N+5, in which the predictions themselves are used to generate predictions. were the best models at N+1 predictions, didn’t perform so good at predictions up to N+5. Finally, we managed to implement an intuitive system, entirely built in R language, which allows the user to access various features, such as analyzing current pandemic data, analyzing the evolution of the pandemic from its beginning to the present day, analyzing forecasts in Portugal and its regions, as well as comparing forecasts.
Description
Keywords
Machine Learning Evolução Previsão Covid-19 Modelos Evolution Forecasting Models