Predicción, análisis y pronóstico de COVID-19 utilizando un modelo de Machine Learning basado en el análisis forecasting sobre series temporales.
Fecha
2021
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad de Guayaquil. Facultad de Ciencias Matemáticas y Físicas. Carrera de Ingeniería en Sistemas Computacionales.
Resumen
El coronavirus (COVID-19), se evidenció por primera vez en la ciudad de Wuhan,
China, el 31 de diciembre de 2019, extendiéndose por el mundo de manera significativa,
llegando a ser oficialmente declarado por la Organización Mundial de la salud como una
pandemia el 11 de marzo del 2020.
En este proyecto se presenta un estudio comparativo de tres modelos de predicción,
tales como: Modelo Autorregresivo AR, modelo de Medias Móviles MA y el modelo Media
Móvil Integrada Autorregresiva Estacional con Regresores Exógenos (SARIMAX), con el fin
de obtener un buen modelo predictivo.
Con el uso de los modelos de series temporales aplicados a un volumen de datos que
previo su ajuste y trasformación, se realizaron varias pruebas de pronósticos con los diferentes
modelos de algoritmos a fin de obtener buen modelo predictivo de COVID-19. Para este
estudio se utilizaron como parámetros de entrada los casos confirmados de contagios.
Este estudio va a permitir ayudar en la toma de decisiones a las autoridades de la salud,
permitiéndole desarrollar estrategias a corto plazo, mediante la utilización de un modelo de
series temporales que permita realizar una predicción futura confiable en periodos cortos que
se ajusten a la serie original para hacer pronósticos.
El proceso realizado para el desarrollo de los modelos de las series temporales son:
Representación gráfica de la serie, transformación del DATASET, ejecución de la función
autocorrelación (F.A.C), ejecución de la función autocorrelación parcial (P.A.C.F), proceso de
entrenamiento, proceso de ajuste, proceso de predicción y visualización de resultados.
The coronavirus (COVID-19), was evidenced for the first time in the city of Wuhan, China, on December 31, 2019, spreading throughout the world significantly, becoming officially declared by the World Health Organization as a pandemic on March 11, 2020. This project presents a comparative study of three prediction models, such as: AR Autoregressive Model, MA Moving Average model and the Seasonal Autoregressive Integrated Moving Average model with Exogenous Regressors (SARIMAX), in order to obtain a good predictive model. With the use of time series models applied to a volume of data that had previously been adjusted and transformed, several forecast tests were performed with the different algorithm models in order to obtain a good predictive model of COVID-19. For this study, confirmed cases of contagion were used as input parameters. This study will help health authorities to make decisions, allowing them to develop short-term strategies, through the use of a time series model that allows making a reliable future prediction in short periods that adjust to the original series for make forecasts. The process carried out for the development of the time series models is: Graphic representation of the series, transformation of the DATASET, execution of the autocorrelation function (FAC), execution of the partial autocorrelation function (PACF), training process, process of adjustment, prediction process and results visualization.
The coronavirus (COVID-19), was evidenced for the first time in the city of Wuhan, China, on December 31, 2019, spreading throughout the world significantly, becoming officially declared by the World Health Organization as a pandemic on March 11, 2020. This project presents a comparative study of three prediction models, such as: AR Autoregressive Model, MA Moving Average model and the Seasonal Autoregressive Integrated Moving Average model with Exogenous Regressors (SARIMAX), in order to obtain a good predictive model. With the use of time series models applied to a volume of data that had previously been adjusted and transformed, several forecast tests were performed with the different algorithm models in order to obtain a good predictive model of COVID-19. For this study, confirmed cases of contagion were used as input parameters. This study will help health authorities to make decisions, allowing them to develop short-term strategies, through the use of a time series model that allows making a reliable future prediction in short periods that adjust to the original series for make forecasts. The process carried out for the development of the time series models is: Graphic representation of the series, transformation of the DATASET, execution of the autocorrelation function (FAC), execution of the partial autocorrelation function (PACF), training process, process of adjustment, prediction process and results visualization.
Descripción
PDF
Palabras clave
COVID-19, Series temporales, Predicción o pronóstico, Machine learning, Análisis estadísticos