medRxiv preprint doi: https://doi.org/10.1101/2020.04.08.20057679.

The copyright holder for this preprint (which was not peer-reviewed)
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-NC-ND 4.0 International license .

COVID-19 Epidemic Analysis using Machine

Learning and Deep Learning Algorithms

Narinder Singh Punn · Sanjay Kumar

Sonbhadra · Sonali Agarwal

Abstract The catastrophic outbreak of Severe Acute Respiratory Syndrome

- Coronavirus (SARS-CoV-2) also known as COVID-2019 has brought the
worldwide threat to the living society. The whole world is putting incredible
efforts to fight against the spread of this deadly disease in terms of infrastruc-
ture, finance, data sources, protective gears, life-risk treatments and several
other resources. The artificial intelligence researchers are focusing their exper-
tise knowledge to develop mathematical models for analyzing this epidemic
situation using nationwide shared data. To contribute towards the well-being
of living society, this article proposes to utilize the machine learning and deep
learning models with the aim for understanding its everyday exponential be-
haviour along with the prediction of future reachability of the COVID-2019
across the nations by utilizing the real-time information from the Johns Hop-
kins dashboard.

Keywords Coronavirus · Machine learning · Deep learning · Prediction

Narinder Singh Punn

IIIT Allahabad, Prayagraj, India, 211015
Tel.: +91-7018466740
E-mail: pse2017002@iiita.ac.in
Sanjay Kumar Sonbhadra
IIIT Allahabad, Prayagraj, India, 211015
Tel.: +91-9827873773
E-mail: rsi2017502@iiita.ac.in
Sonali Agarwal
IIIT Allahabad, Prayagraj, India, 211015
Tel.: +91-9415647042
E-mail: sonali@iiita.ac.in
medRxiv preprint doi: https://doi.org/10.1101/2020.04.08.20057679. The copyright holder for this preprint (which was not peer-reviewed)
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-NC-ND 4.0 International license .

2 Narinder Singh Punn et al.

Fig. 1 Electron microscopy imaging of COVID-19 representing morphological features at

varying magnification levels [4].

1 Introduction

Coronaviruses are a large family of viruses that can cause severe illness to the
human being. The first known severe epidemic is Severe Acute Respiratory
Syndrome (SARS) occurred in 2003, whereas the second outbreak of severe
illness began in 2012 in Saudi Arabia with the Middle East Respiratory Syn-
drome (MERS). The current outbreak of illness due to coronavirus is reported
in late December 2019. This new virus is very contagious and has quickly
spread globally. On January 30, 2020, the World Health Organization (WHO)
declared this outbreak a Public Health Emergency of International Concern
(PHEIC) as it had spread to 18 countries. On Feb 11, 2020, WHO named this
“COVID-19”. On March 11, as the number of COVID-19 cases has increased
thirteen times apart from China with more than 118,000 cases in 114 countries
and over 4,000 deaths, WHO declared this a pandemic.
As the outbreak of the COVID-19 has become a worldwide pandemic, the
real-time analyses of epidemiological data are needed to prepare the society
with better action plans against the disease. Since the birth of novel COVID-
19 [1], the world is restlessly fighting with its cause. As of April 1, 2020, based
on the globally shared live data by the Johns Hopkins dashboard, worldwide
there are 932,605 confirmed cases, out of which 193,177 are recovered and
46,809 lost their lives [2]. COVID-19 belongs to the family of the SARS-CoV
and MERS-CoV, where it begins with the initial level symptoms of the com-
mon cold to severe level of respiratory diseases causing difficulty in breathing,
tiredness, fever, and dry cough [3]. Prasad et al. [4] observed that the identifica-
tion of the virus can be improved by imaging using immunoelectron microscopy
techniques [5]. Figure 1 shows the typical structure of COVID-19 virus, from
the throat swab of the first Indian laboratory-confirmed case, captured using
sample electron microscopy imaging. Till date, detailed morphology and ul-
trastructure of this virus remain incompletely understood, and there are no
medRxiv preprint doi: https://doi.org/10.1101/2020.04.08.20057679. The copyright holder for this preprint (which was not peer-reviewed)
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-NC-ND 4.0 International license .

COVID-19 Epidemic Analysis 3

Fig. 2 Timeline of COVID-19 across the nations.

specific vaccines or treatments for COVID-19. However, many ongoing clinical

trials are evaluating potential treatments.

1.1 COVID-19 transmission stages

As per the WHO reports [5], the pandemic situation is classified into four
stages. The first stage begins with the cases reported for the people who trav-
elled in already affected regions, whereas in the second stage, cases are reported
locally among family, friends and others who came into contact with the per-
son arriving from the affected regions. At this point the affected people are
traceable. Later, the third stage makes the situation even worse as the trans-
mission source becomes untraceable and spreads across the individuals who
neither have any travel history nor came into contact with the affected per-
son. This situation demands immediate lockdown across the nation to reduce
the social contacts among individuals and control the rate of transmission.
Figure 2, presents the outbreak of COVID-19 across the nations with the high
number of confirmed cases such as US, Italy, Spain, and China based on the
WHO reports [3], whereas figure 3 illustrates the rising characteristic of the
number of confirmed, death, and recovered cases in these nations between the
period January 22, 2020, to April 1, 2020, taken from Johns Hopkins live dash-
board. The worst of all, stage 4 beings when the transmission becomes endemic
and uncontrollable. Figure 4 presents the complete stages of the COVID-19
epidemic. Until now, several countries have entered stage 4. China is the first
nation that experienced the stage 4 of the COVID-19 transmission. Though,
medRxiv preprint doi: https://doi.org/10.1101/2020.04.08.20057679. The copyright holder for this preprint (which was not peer-reviewed)
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-NC-ND 4.0 International license .

4 Narinder Singh Punn et al.

Fig. 3 Epidemic status of the COVID-19 across nations having the highest number of
confirmed cases since January 22, 2020.

Fig. 4 COVID-19: complete transmission stages.

it is claimed that the origin of the virus is Wuhan, China [6]; it affects the
other developed countries (USA, Italy, Spain, Britain, etc.). These countries
are now in stage 4 of the transmission and facing more number of infections
and deaths compared to China. In the case of China, it is observed that expo-
nential growth of the confirmed cases reaches the saturation stage where the
number of cases stopped growing. This follows from the fact that the number
of susceptible people, which are exposed to virus, are dramatically reduced.
This was made possible due to the reduced social contact among people by
segregating the infected individuals in quarantine and a complete lockdown pe-
riod was initiated by the Chinese government, thereby reducing the possibility
of further spread.
Machine learning algorithms play an important role in epidemic analysis
and forecasting [7], [8]. In the presence of massive epidemic data, the machine
learning techniques help to find the epidemic patterns so that the early action
can be planned to stop the spread of the virus. In this research, machine learn-
medRxiv preprint doi: https://doi.org/10.1101/2020.04.08.20057679. The copyright holder for this preprint (which was not peer-reviewed)
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-NC-ND 4.0 International license .

COVID-19 Epidemic Analysis 5

ing and deep learning models are used to observe everyday behaviour along
with the prediction of future reachability of the COVID-2019 across the nation
by utilizing the real-time information from the Johns Hopkins dashboard.
Rest of the paper is organised as follows: section 2 contains a brief discus-
sion on existing machine learning based approaches, whereas section 3 talks
about the dataset used for experiments. Epidemic analysis and experimen-
tal strategy are discussed in section 4 and 5 respectively, whereas results are
discussed in section 6. Concluding remarks and future work are discussed in
section 6.

2 Related work

Since the last decade, digital technologies are playing critical roles in major
health sector problems including disease prevention, the present worldwide
health emergency also seeking technological support to tackle COVID-2019.
In a paper, authors highlighted the possible applications of trending digital
technologies such as internet of things (IoT), big-data analytics, artificial in-
telligence (AI), deep learning and blockchain technology to develop strategies
for monitoring, detection and prevention of epidemic; and also to identify the
impact of the epidemic to the healthcare sector [9]. In a research work pro-
posed by Benvenuto et al. [10], authors proposed an autoregressive integrated
moving average (ARIMA) model to predict the spread of COVID-2019. In this
paper, the author forecasted the various parameters for the next 2 days based
on the study about the prevalence and incidence of the COVID-2019. This
research work also demonstrates the correlogram and ARIMA forecast graph
for the epidemic incidence and prevalence.
Deb et al. [11] proposed a time series method to analyze incidence pat-
tern and the estimated reproduction number of COVID-19 outbreak. They
performed statistical analysis to explore the trends of the outbreak to high-
light the present epidemiological stage of a region so that various policies can
be identified to address COVID-19 pandemic in different countries cite11. As
per the present situation, it is essential to understand the early spread pat-
terns of the infection to plan and control the effective safety measures. In this
direction, Kucharski et al. proposed a scientific model of critical SARS-CoV-
2 transmission by using different datasets to study the COVID-19 outbreak
inside and outside Wuhan. With this, they explored the possible spread of
disease outbreak outside Wuhan [12].
Recently there are several studies conducted on the epidemiological out-
break of COVID-19 using exploratory data analysis (EDA) based on various
available datasets. The studies mainly focus on the occurrence of confirmed,
death, and recovered cases in Wuhan and the rest of the world to understand
the suspected threats and subsequent planning of containment activities [13].
Lauer et al., in their research work raised the issue of the criticality of the
incubation period for COVID-19. They studied 181 confirmed cases and iden-
tified that the incubation period may vary from 5 days to 14 days and based
medRxiv preprint doi: https://doi.org/10.1101/2020.04.08.20057679. The copyright holder for this preprint (which was not peer-reviewed)
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-NC-ND 4.0 International license .

6 Narinder Singh Punn et al.

on this better surveillance and control activities can be planned [14]. In recent
research work, Singer analyzed data of 25 infected counties to follow short term
predictions about the COVID 2019 outbreak. The research highlighted that
the location-specific rate of disease spread follows either steady or explosive
power-law growth with different scaling exponents. With this understanding,
the authors analyzed the impact of lockdown in various parts of the world [15].
Based on the above literature, it is evident that sufficient work is available on
exploratory data analysis to understand the existing trend of the epidemic but
still there is a lot of scopes to develop and test efficient machine learning based
prediction models so that proactive strategies could be identified to cater the
immediate needs.

3 Dataset description

The day to day prevalence data of COVID-2019 from January 22, 2020, to
April 1, 2020, were retrieved from the official repository of Johns Hopkins
University. The dataset consists of daily case reports and daily time series
summary tables. In the present study, we have taken time-series summary
tables in CSV format having three tables for confirmed, death and recovered
cases of COVID-2019 with six attributes i.e. province/state, country/region,
last update, confirmed, death and recovered cases, where the update frequency
of the dataset is once in a day [2]. Figure 5 presents the COVID-19 confirmed,
recovered, and death cases distribution across the world since the time data
was recorded. It is easy to observe the exponential growth of the spread which
needs to be controlled.

4 Epidemic Analysis

The COVID-19 spread has brought the world under the brink of loss of hu-
man lives due to which it is of utmost importance to analyze the transmission
growth at the earliest and forecast the forthcoming possibilities of the trans-
mission. With this objective, state-of-the-art mathematical models are adopted
based on machine learning such as support vector regression (SVR) [16] and
polynomial regression (PR) [17], and deep learning regression models such as
a standard deep neural network (DNN) and recurrent neural networks (RNN)
using long short-term memory (LSTM) cells [18]. Machine learning and deep
learning approaches are implemented using the python library “sklearn” and
“keras” respectively, to predict the total number of confirmed, recovered, and
death cases worldwide. The prediction will allow undertaking the necessary de-
cisions based on transmission growth such as increasing the lockdown period,
executing the sanitation procedure, providing the everyday resources, etc.
medRxiv preprint doi: https://doi.org/10.1101/2020.04.08.20057679. The copyright holder for this preprint (which was not peer-reviewed)
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-NC-ND 4.0 International license .

COVID-19 Epidemic Analysis 7

Fig. 5 Distribution of COVID-19 cases across the globe.

5 Training and testing

The regression approaches for epidemic analysis are trained and tested on real-
time data [2] using the number of confirmed, recovered, and death cases as the
label for the corresponding day. With extensive experiments, machine learning
approaches are implemented with the polynomial kernel of degree 6 and other
coefficient values as gamma=0.01, epsilon=1, and C=0.1. The standard DNN
consists of a dense input layer with 128 neurons, three hidden dense layers
with 256 neurons and output layer is consists of a single neuron whereas the
RNN, consists of three stacks of LSTM layers having 64 neurons combined
with 10% dropout to avoid the overfitting problem and final output layer
with a single neuron. The mean squared error (MSE) is the most widely used
objective function and root mean square error (RMSE) as a metric function
for evaluating the regression models. The MSE loss can be computed by using
equation 1.

N
1 X
L(y, ŷ) = (y − ŷ) (1)
N i=0

where y indicates the original value, ŷ indicates the predicted value, and
N is the number of samples predicted.
Due to the limited availability of the data, the trained models are validated
against the training data and then utilized to forecast the number of confirmed,
recovered, and death cases for the next 10 days.

6 Results and discussion

The discussed machine learning and deep learning approaches output the pos-
sible number of cases for the next 10 days across the world. Figure 6 illustrates
the predicted trend of the COVID-19 using SVR, PR, DNN, and LSTM with
worldwide data. Table 1 indicates the RMSE score of the approaches com-
puted against the available number of COVID-19 cases. It is also observed that
training of LSTM model is heavily dependent on the deviation in the values,
with the fact that larger the deviation more the time it takes to train. Hence,
the number of cases were scaled using minmax scaler to fit the LSTM model
and later the predicted cases were rescaled to the original range using invert
medRxiv preprint doi: https://doi.org/10.1101/2020.04.08.20057679. The copyright holder for this preprint (which was not peer-reviewed)
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-NC-ND 4.0 International license .

8 Narinder Singh Punn et al.

Fig. 6 COVID-19 worldwide epidemic analysis using SVR, DNN, LSTM, and PR.

minmax transform from “sklearn” python library. Among these approaches,

the visual representation of the prediction from figure 6 and RMSE score as
highlighted in table 1 confirms the PR approach as the best fit to follow the
growing trend.

Table 1 Comparative results.

Model RMSE 10-04-2020 (Forecast)
Confirmed Deaths Recovered Confirmed Deaths Recovered
SVR 27456.47 1360.47 16762.15 1970860 93112 422740
DNN 163335.65 8554.55 25415.03 211213 7419 93476
LSTM 15647.64 1076.06 4092.01 1002543 50362 206916
PR 455.92 117.94 809.71 2156540 124021 557125

7 Conclusion

The world is under the grasp of SARS-CoV2 (COVID-19) virus. Early pre-
diction of the transmission can help to take necessary actions. This article
proposed to utilize the machine learning and deep learning models for epi-
demic analysis using data from Johns Hopkins dashboard. The results show
that polynomial regression (PR) yielded a minimum root mean square error
medRxiv preprint doi: https://doi.org/10.1101/2020.04.08.20057679. The copyright holder for this preprint (which was not peer-reviewed)
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-NC-ND 4.0 International license .

COVID-19 Epidemic Analysis 9

(RMSE) score over other approaches in forecasting the COVID-19 transmis-

sion. However, if the spread follows the predicted trend of the PR model then
it would lead to huge loss of lives as it presents the exponential growth of the
transmission worldwide. As observed from China, this growth of the COVID-19
can be reduced and quenched by reducing the number of susceptible individ-
uals from the infected individuals. This is achievable by becoming unsocial
and following the lockdown initiative with discipline. The study can further
be extended to utilize other machine learning and deep learning models.

Compliance with Ethical Standards

Conflict of Interest

The authors have no conflict of interest to declare.

Ethical approval

This article does not contain any studies with human participants or animals
performed by any of the authors.

References

1. Stoecklin, Sibylle Bernard, Patrick Rolland, Yassoungo Silue, Alexandra Mailles, Chris-
tine Campese, Anne Simondon, Matthieu Mechain et al. ”First cases of coronavirus disease
2019 (COVID-19) in France: surveillance, investigations and control measures, January
2020.” Eurosurveillance 25, no. 6 (2020)
2. Github repository. 2019 Novel Coronavirus COVID-19 (2019-nCoV) Data
Repository by Johns Hopkins CSSE. Retrieved March 31, 2020 from
https://github.com/CSSEGISandData/COVID-19.
3. WHO coronaviruses (COVID-19). Retrieved March 30, 2020 from
https://www.who.int/emergencies/diseases/novel-coronavirus-2019.
4. Prasad, Sharda, Varsha Potdar, Sarah Cherian, Priya Abraham, and Atanu Basu. ”Trans-
mission electron microscopy imaging of SARS-CoV-2.” The Indian journal of medical re-
search (2020).
5. Leppard, Gary G. ”Evaluation of electron microscope techniques for the description of
aquatic colloids.” In Environmental particles, pp. 231-289. CRC Press, 2019.
6. Jia, L., Li, K., Jiang, Y., Guo, X. (2020). Prediction and analysis of Coronavirus Disease
2019. arXiv preprint arXiv:2003.05447.
7. Kalipe, G., Gautham, V., Behera, R. K. (2018, December). Predicting malarial outbreak
using Machine Learning and Deep Learning approach: A review and analysis. In 2018
International Conference on Information Technology (ICIT) (pp. 33-38). IEEE.
8. Singh, R., Singh, R., Bhatia, A. (2018). Sentiment analysis using Machine Learning
technique to predict outbreaks and epidemics. Int. J. Adv. Sci. Res, 3(2), 19-24.
9. Ting, Daniel Shu Wei, Lawrence Carin, Victor Dzau, and Tien Y. Wong. ”Digital tech-
nology and COVID-19.” Nature Medicine (2020): 1-3.
10. Benvenuto, Domenico, Marta Giovanetti, Lazzaro Vassallo, Silvia Angeletti, and Mas-
simo Ciccozzi. ”Application of the ARIMA model on the COVID-2019 epidemic dataset.”
Data in brief (2020): 105340.
medRxiv preprint doi: https://doi.org/10.1101/2020.04.08.20057679. The copyright holder for this preprint (which was not peer-reviewed)
is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.
It is made available under a CC-BY-NC-ND 4.0 International license .

10 Narinder Singh Punn et al.

11. Deb, Soudeep, and Manidipa Majumdar. ”A time series method to analyze in-
cidence pattern and estimate reproduction number of COVID-19.” arXiv preprint
arXiv:2003.10655 (2020).
12. Kucharski, Adam J., Timothy W. Russell, Charlie Diamond, Yang Liu, John Edmunds,
Sebastian Funk, Rosalind M. Eggo et al. ”Early dynamics of transmission and control of
COVID-19: a mathematical modelling study.” The lancet infectious diseases (2020).
13. Dey, Samrat Kumar, Md Mahbubur Rahman, Umme Raihan Siddiqi, and Arpita
Howlader. ”Analyzing the Epidemiological Outbreak of COVID-19: A Visual Exploratory
Data Analysis (EDA) Approach.” Journal of Medical Virology (2020).
14. Lauer, Stephen A., Kyra H. Grantz, Qifang Bi, Forrest K. Jones, Qulu Zheng, Hannah
R. Meredith, Andrew S. Azman, Nicholas G. Reich, and Justin Lessler. ”The incubation
period of coronavirus disease 2019 (COVID-19) from publicly reported confirmed cases:
estimation and application.” Annals of internal medicine (2020).
15. Singer, H. M. ”Short-term predictions of country-specific Covid-19 infection rates based
on power law scaling exponents.” arXiv preprint arXiv:2003.11997 (2020).
16. Awad, Mariette, and Rahul Khanna. ”Support vector regression.” In Efficient Learning
Machines, pp. 67-80. Apress, Berkeley, CA, 2015.
17. De Castro, Yohann, Fabrice Gamboa, Didier Henrion, Roxana Hess, and Jean-Bernard
Lasserre. ”Approximate optimal designs for multivariate polynomial regression.” The An-
nals of Statistics 47, no. 1 (2019): 127-155.
18. Greff, Klaus, Rupesh K. Srivastava, Jan Koutnı́k, Bas R. Steunebrink, and Jürgen
Schmidhuber. ”LSTM: A search space odyssey.” IEEE transactions on neural networks
and learning systems 28, no. 10 (2016): 2222-2232.