Article Forecasts for confirmed COVID-19 cases using CNN, ARIMA and Exponential Smoothing Juan Frausto-Solís 1,†,‡ , Lucía J. Hernández-González 1,‡ , Juan J. González-Barbosa 1,†,‡ *, Edgar Román-Rangel 1,†, , Juan Paulo Sánchez-Hernández 2,† Citation: Frausto-Solís, J.; Hernández-González, L.; González-Barbosa, J.; Román-Rangel, E.; Sánchez-Hernández, J. Forecasts for confirmed COVID-19 cases using CNN, ARIMA and Exponential Smoothing. Journal Not Specified 2021, 1, 0. https://doi.org/ Received: Accepted: Published: Publisher’s Note: MDPI stays neu- tral with regard to jurisdictional claims in published maps and insti- tutional affiliations. Copyright: © 2021 by the authors. Submitted to Journal Not Specified for possible open access publication under the terms and conditions of the Creative Commons Attri- bution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). 1 División de Estudios de Posgrado e Investigación. Tecnológico Nacional de México/Instituto Tecnológico de Ciudad Madero, Av. 1o. de Mayo esq. Sor Juana Inés de la Cruz s/n, Col. Los Mangos C.P.89440 Cd. Madero Tamaulipas, México; luciajaneth.hernandez@gmail.com (L.J.H.-G.), juan.frausto@gmail.com (J.F.-S.), jjgonzalezbarbosa@hotmail.com (J.J.G.-B.) 2 Dirección de Informática, Electrónica y Telecomunicaciones, Universidad Politécnica del Estado de Morelos, Boulevard Cuauhnáhuac 566, Col. Lomas del Texcal, Jiutepec, Morelos. C.P.62550; juan.paulosh@upemor.edu.mx (J.P.S.-H.) 3 Departamento Académico de Sistemas Digitales, Instituto Tecnológico Autónomo de México 01080 Ciudad de México; edgar.roman@itam.mx (E.R-R.) * Correspondence: frausto208008@gmail.com (J.J.G.-B.) † Current address: Affiliation 3 ‡ These authors contributed equally to this work. Abstract: On a global level, there has been a health contingency due to the COVID-19 disease. 1 This disease has left millions of infected people and its spread dramatically increased. This 2 study proposes a new method based on Convolutional Neural Network (CNN) and temporal 3 Component Transformation (CT) called CNN-CT. This method is applied to confirmed cases of 4 COVID-19 in the United States and Mexico. The CT changes daily predictions and observations 5 to weekly components and vice versa. Besides, CNN-CT adjusts the predictions made by CNN 6 using ARIMA and Exponential Smoothing (ES) methods. This combination of strategies provides 7 better predictions than any of the individual methods by themselves. In this paper, we present 8 the mathematical formulation for this strategy. Our experiments encompass the fine-tuning of 9 the parameters of the algorithms. We compared the best hybrid methods obtained with CNN-CT 10 versus the individual CNN, ARIMA, and ES methods. The results show that our hybrid method 11 surpasses the performance of the individual methods. Several error metrics are used to evaluate 12 the performance of our methods, obtaining improvements ranging from 1.76% to 46%, with respect 13 to individual methods. 14 Keywords: forecast, COVID-19, CNN, ARIMA, Exponential Smoothing. 15 1. Introduction 16 Coronaviruses are a large family of viruses characterized by having crown-shaped 17 spikes on their surface. Nowadays, they are seven identified types of coronaviruses that 18 can be transmitted among humans. The most dangerous coronaviruses known until 19 recent years are MERS-CoV and SARS-CoV, and they had provoked severe diseases, 20 such as MERS and SARS, in 2003 and 2012, respectively [1]. However, at the end of 2019, 21 in Wuhan, China, the new epidemiological outbreak of COVID-19 emerged; it caused 22 the new coronavirus called SARS-CoV2. 23 The importance of mathematical models and algorithms to analyze this disease has 24 grown because they allow us to find patterns, predictions and understand fluctuations. 25 Epidemiological models can be classified into two groups [2]: 26 • Dynamic Models. They are ancient models that usually divide the population into 27 three subsets known as compartments: susceptible, infected, and recovered. SIR 28 Version February 27, 2021 submitted to Journal Not Specified https://www.mdpi.com/journal/notspecified