Predicting Covid-19 Cases using CNN Model Paul Menounga Mbilong 1 a , Asmae El Kassiri 1 , Fatima-Zahra Belouadha 1 and El Bhiri Brahim 2 1 AMIPS Research Team, Ecole Mohammadia d’Ingénieurs, Mohammed V University in Rabat 2 SMARTLIB, Ecole Marocaine des Sciences de l'Ingénieur Rabat, Morocco Keywords: The COVID-19, CNN, LSTM, Deep Learning, Time-series forecasting. Abstract: The prediction of COVD-19 confirmed cases is a complex time-series problem. In the literature, Long Short Time Memory (LSTM) has proven its efficiency to resolve issues related to the time series problems. However, Convolution neural network (CNN) did not been widely used in this aim and is considered as more suitable for imaging processing. Therefore, in this paper, we use it to predict COVID-19 cases and compared it with LSTM in the context of Morocco during the period of confinement. The obtained results which we present and discuss in this article are very promising. 1 INTRODUCTION Reported first in December 2019, in Wuhan, China, Coronavirus 2019 (COVID-19), caused by the strain of coronavirus named “Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2)”, has become now, according to the World Health Organization WHO, an international epidemic having caused 17 660 523 infected cases, and 680 894 deaths on August 03, 2020 (Organization, 2020). Since the first cases were reported, multiple academic and medical works have explored various approaches to find a solution for the COVID-19 epidemic in different search areas including the Machine Learning (ML). Some ML, and especially Deep Learning (DL), work interested in COVID-19 are oriented to help medical staff to efficiently diagnostic infected people (Singh, Kumar, Vaishali, & Kaur, Coronavirus (COVID-19) Classification using CT Images by Machine Learning Methods, 2020) (Wang, et al., 2020) (Gozes, et al., 2020) (Metsky, Freije, Kosoko-Thoroddsen, & Myhrvold, 2020), and others are directed to analyze the pandemic situation like predicting new patients to contribute to local hospital arrangement (Alimadadi, et al., 2020) (Bouhamed, 2020) (Chimmula & Zhang, 2020) (Pinter, Felde, MOSAVI, Ghamisi, & Gloaguen, 2020). In general, a lot of work applied ML and DL to predict COVID-19 new cases use time-series a https://orcid.org/0000-0001-5464-4803 forecasting models (Chimmula & Zhang, 2020) (Zeroual, Harrou, Dairid, & Sun, 2020) (Azarafza, Azarafza, & Tanha, 2020) (Punn, Sonbhadra, & Agarwal, 2020). Actually, Recurrent Neural Networks (RNN) architecture are adopted by several searchers thanks to their capacity in handling time- dependent datasets, and hence they were widely used in the context of predicting COVID-19 new cases. Moreover, other Neural Networks (NN) architectures were used for this objective. The objective of this work is to use Convolution Neural Network (CNN) architecture in time-series forecasting to predict the COVID-19 new cases in Morocco. We compare our results with a Long Short- Term Memory (LSTM) neural network explored in the same context by another work conducted by our research team. This paper is constituted of five sections. The second one explores the related work, which is relative to CNN architectures and their use in Time- series forecasting, especially in the context of COVID-19. The third section is about Materials and Method used to perform our experimental studies. We present and discuss the results in the fourth section, and finally we conclude in the fifth section.