Journal of Advances in Mathematics and Computer Science

36(3): 97-117, 2021; Article no.JAMCS.67121

ISSN: 2456-9968
(Past name: British Journal of Mathematics & Computer Science, Past ISSN: 2231-0851)

SIR Model Parameters Estimation with COVID-19

Data
∗
Nilson C. Roberty1 and Lucas S. F. de Araujo2
1 Nuclear Engineering Program, Coppe, Universidade Federal do Rio de Janeiro, Av. Horócio

Macedo, 2030, Bloco G, Sala 206, Centro de Tecnologia, Cidade Universitária, Ilha do Funda̋o,
21941-914, Rio de Janeiro, RJ, Brasil.
2 Nuclear Engineering Department, Poli, Universidade Federal do Rio de Janeiro, Av. Horócio

Macedo, 2030, Bloco G, Sala 206, Centro de Tecnologia, Cidade Universitária, Ilha do Funda̋o,
21941-914, Rio de Janeiro, RJ, Brasil.

Authors’ contributions

This work was carried out in collaboration between both authors. Both authors read and approved
the final manuscript.

Article Information
DOI: 10.9734/JAMCS/2021/v36i330349
Editor(s):
(1) Dr. Junjie Chen, University of Texas at Arlington, USA.
(2) Dr. Rodica Luca, Gh. Asachi Technical University, Romania.
(3) Dr. Paul Bracken, The University of Texas RGV, USA.
Reviewers:
(1) Adil Osman Mageet, Ajman University, UAE.
(2) Mojeeb Al-Rahman El-Nor Osman El-Haj, International University of Africa, Sudan.
(3) Li Wang, Ningxia University, China.
(4) Theodore G. Lewis Naval, Postgraduate School Monterey, USA.
(5) Samson Wangila Wanyonyi, University of Eldoret, Kenya.
(6) Vipin Tiwari, Kumaun University, India.
Complete Peer review History: http://www.sdiarticle4.com/review-history/67121

Received: 14 February 2021

Accepted: 18 April 2021
Original Research Article Published: 24 April 2021

Abstract
Based on the SIR model that divides the population into susceptible, infected and removed
individuals, data about the evolution of the pandemic compiled by the Johns Hopkins University
Center for Systems Science and Engineering (JHUCSSE) are integrated into the numerical system
solution. The system parameters Rate of Contact β, Basic Reproduction Number R0 and Removal

*Corresponding author: E-mail: [email protected];

 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Rate γ, also named Rate of Decay, are determined according to a ridge regression approach
and a mobile statistical scheme with different averages. Data is automatically downloaded from
https://raw.githubusercontent.com/CSSEGISandData/COVID-19. The main Python libraries
used are Numpy, Pandas, Skit-Learn, Requests and Urllib.

Keywords: COVID-19; Python3; Inverse Problems; Ridge Regression; ODE; Fixed Points.

2010 Mathematics Subject Classification: 34A55, 62J07, 34XX, 37C25.

1 Introduction
The simplest model capable of describing the three main states of the phenomenology associated
with ’COVID-19 infection’ is the SIR model ([1],[2]). These three letters designate the three
compartments grouping the population affected by the disease. The susceptible group (S) correspond
to all people that can be in contact with the virus and contract the disease. By contracting the
disease, susceptible people will become part of the temporary group of infected people (I). After the
characteristic time of the evolution of the disease, those infected individuals migrate to the removed
group (R) that counts both the people who get recovered and those who died due to the severity of
clinical conditions. This model is based on two parameters traditionally denoted as β and γ, which
contemplate two basic types of fundamental mechanisms found in the phenomena of reaction and
decay ([3]).

The emergence of corona virus epidemic at the end of 2019 has led the world to an unprecedented
crisis. The last infectious disease of this size had occurred a century ago, exponentially increasing
the social consequences that followed the first world war. At this time, the importance of scientific
treatment for pests and epidemics emerges in the wake of these consequences and the SIR model
appears as a simple and adequate tool for the quantification of this type of problem.Important
books on mathematical modeling of populations in biology and epidemiology have been published,
we quote ([4]). As a nonlinear system of ordinary differential equations, the SIR model has an
important position as a pure and applied mathematics subject. Since in the modelling of real
infections the two main parameters depends on social contact and therapeutics procedures, and
consequently are strongly determined by public politics, they will change with time and the SIR
model will behave as a non autonomous dynamical systems ([5]). Also, very disputed political
decisions based on different interests and beliefs of the authorities increase their stochastic nature.
Interesting works has been done before the present outbreak, we cite only some([6],[7]). More
recently([8]) investigate analytical features of the SIR model and their applications to COVID-19,
and ([9]) comments those results. In a more recent work ([10]) he presents an analytical parameter
estimation of COVID-19 data with a different methodology from the one used in this work. Other
interesting recent paper treat the COVID-19 data forecasting problem ([11]). We will not treat this
forecasting problem in this first work.

In section (2) we describe the SIR model and we give a brief description of the parameters. In
subsection (2.1) we define the Basic Reproduction Number. In subsection (2.2) we inform more
about the data spreadsheets which were used. In Section (3) we derive a fix point equation and
define the concept of Local Pandemic Cycle. Section (4) describes how Python3.7 is used to extract
the data from JHUCSSE website and produce our results. Section (5) describes the numerical
implementation of the Tychonoff regularization for parameters reconstruction. The results are
presented in section (6). Some concluding remarks are presented in section (7).

98
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

2 The SIR Model

The equations of the SIR model with time-dependent β(t) and γ(t) are:


 dt = −β(t) N I(t)
dS(t) S(t)

dI(t)
= β(t) N I(t) − γ(t)I(t)
S(t)
(2.1)
 dt
 dR(t) = γ(t)I(t)
dt

with the initial condition (S(t0 ), I(t0 ), R(t0 )) = (S0 , I0 , R0 ).

Brazil

Confirmed Confirmed
Removed 10000 Removed
10000 Infected Infected

Number of Cases using reconstructed parameters(x1000)

Number of Cases from JHU-CCSE data(x1000)

8000 8000

6000
6000

4000
4000

2000
2000

0
0
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
Days Days

Fig. 1. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The
Brazilian Data Case

The functions S(t), I(t), R(t) respectively represent the susceptible, infected and removed (recovered
plus death) number of people at any given time. The shape of these equations allows us to make a
parallel with the kinetic nuclear reaction equations which involves reactions and radioactive decay
([3]).

The parameter N can be interpreted as the total number of people involved in the epidemic outbreak.
The sum of the three equations (2.1) establishes N as a time invariant

S(t) + I(t) + R(t) = N (2.2)

The first parameter is the Contact Rate, β. It establishes the rate of contact per person in a
S
population with N individuals. Since the susceptible fraction, N , represents the probability of
S
finding a susceptible individual in the population, an infected person have in average β N contacts
with susceptible people.

The second parameter is the removal rate, γ, which controls the decay rate from the infected state
to the removed state. These type of parameters are also found in nuclear reaction phenomenology.
The probability of an individual leave the infected condition in an infinitesimal increment of time
δτ is γδτ . Therefore, the probability of remaining in the infected condition is 1 − γδτ . Thus, in
the classical limit, we have the quantification of the probability that in a total time τ an individual
remains in the condition of infected

99
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

τ
lim (1 − γδτ ) δτ = e−γτ (2.3)
δτ →0

This behavior is typical in the nuclear radioactive decay phenomena. Meanwhile in the context of
the epidemic outbreak it is observed that individuals are more likely to recover or die in a few days
after contracting the infection, this simplification is systematically adopted and useful.
The population number N is normally very large when compared with the confirmed number of
S
cases directly involved in the epidemic, at least in its beginning. Thus, the hypothesis N ≃ 1 makes
the model linear. It is appropriate to define the Effective Value of the Rate of Contact βef .

S
βef = β (2.4)
N
The average number of contacts of an infected individual in an infinitesimal time δτ is βef δτ . Thus
1 − βef δτ represents the probability that the infection will not be transmitted in the time interval
τ
δτ . Since this interval has δτ sub-intervals, an infected individual remains unlikely to infect others
with probability:
τ
lim (1 − βef δτ ) δτ = e−βef τ
δτ →0

Consequently the probability of the transmission is 1 − e−βef τ ([1]).

2.1 The basic reproduction number R0

An individual who remains infected for a time τ will have in average βτ contacts and will infect
susceptible individuals who have had contact with him ([1]). The probability that this individual
remain infected in this interval of time is given by the equation(2.3), and the probability of finishing
the period of infection at the moment τ will be γe−γτ for all time τ . In this circumstance, the average
Basic Reproduction Number of the infection will be:
∫ ∞
β
R0 = βτ γe−γτ dτ = . (2.5)
0 γ

The Basic Reproduction Number R0 := βγ has a direct influence on local evolution of the pandemic.
As it will be seen in the next sections, it controls the evolution of the epidemic cycle, as shown
in the figures 1 and 2. The control of the Contact Rate between individuals can reduce the basic
reproduction number to levels low enough to prevent an epidemic outbreak of great proportions.
All this depends of the society’s ability to reduce its values to below 1. Different countries are at
different stages in the development of the pandemic.
Effective health authorities efforts to contain the outbreak will introduce a trend towards a reduction
in the basic number of reproduction by decreasing the Contact Rate β. Further study can be
conducted also with network science ([1]).

2.2 The COVID-19 data

On the website https://data.humdata.org/dataset/novel-coronavirus-2019-ncov-cases, ([12]), it is
available 3 time series with information used for analysis on the cumulative number of cases of
Covid19 registered as confirmed, recovered and deaths for several countries. For the automatic
download there is an alternative GitHub link in the website https://raw.githubusercontent.com
which is synchronized with the JHUCSSE data. For some countries, there is discrimination against
provinces and regions. The combination of the numerical solution of the dynamic SIR equations
solved as a deterministic system of ordinary differential equations with parameters β , γ and N
allows a temporal characterization of the system states.

100
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Those results can be compared with the observational data in the JHUCSSE worksheets:

{conf irmed.csv, recovered.csv, death.csv} . (2.6)

In this way, we establish an Observational-Evolution model that can be seen both as a Statistical
filtering problem for discrete times and also as a problem of deterministic extrapolation with
constant parameters in a long-range time. In this first work we will not explore the extrapolation
([13]),([14]).

3 The Existence of the Outbreak in the SIR Model

The first system equation (2.1) with time constant parameters β and γ can be solve explicitly giving
∫
β t I(τ )dτ ∫t
S(t) = S(0)exp(− 0 N ). By integrating the third equation R(t) − R(0) = γ 0
I(τ )dτ ). The
combination of these equations gives
R(t) − R(0)
S(t) = S(0)exp(−R0 ) (3.1)
N
The replacement of (3.1) in the equation (2.2) provides the following equation with normalized
variables r(t) := R(t)
N
,i(t) := I(t)
N
and S(t) := S(t)
N
for the removal rate

dr(t)
= γ(1 − r(t) − (1 − i(0)))e−R0 r(t) ). (3.2)
dt
This equation (3.2) is equivalent to the SIR model in the case of constant parameters.

New Zealand
Population = 4.886 millions
Infected
2500 Confirmed
Removed
2000

1500
Numbers

1000

500

0
Jan,22,20 May,1,20 Aug,9,20 Nov,17,20 Feb,25,21
Days

Fig. 2. Local pandemic cycle in New Zealand. The Local Pandemic Cycle is defined
between March and June 2020 by the area delimited by the curves confirmed and
removed, respectively

We can see that the rates drdt

di
and dt are both null at the beginning and at the end of the outbreak.
The evolution of the curves that represents the number of confirmed cases c(t) = r(t) + i(t) and
that represents the number of removed cases r(t) delimits a cycle of the state evolution process.
The area between the curves can be seen as a measure of the size of the outbreak. This region of
the process will be called as Local Pandemic Cycle. And, of course, after a complete cycle, nothing

101
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

prevents a new outbreak from starting after the final state of the first outbreak, starting a new
local pandemic cycle. In the Fig. 2 we can see the formation of a complete first cycle and the
adequate control of the disease. The rates dr
dt
and dc
dt
are both null at the beginning and at the end
of the outbreak between March and June 2020. At this period of time, only the residual number of
infected people is observed.
So, the Fig. 2 shows the first New Zealand epidemic cycle in which the result of controlling the
R0 parameter to values under 1 led to a low value of rϵ ≃ 4886000
310
≃ 6.3410−5 which measures the
infected fraction of the population.
Since at the extremes of a cycle the derivatives are zero, we get the following fixed point equation,
which is also known as giant component size equation,
−β r
r = 1 − (1 − ϵ)e γ (3.3)
The numerical solution of this fix point equation is done with the Newton method.
When the influence of the initial disturbance ϵ = IN0 ≃ 0 is neglected, this equation will always
have a zero solution, r(t) = 0, and can, depending of the value R0 = βγ have another non-null
solution([1]). In the case of R0 = 2, the non-null solution is r ≃ 0.7968121301668857, showing that
almost 80% of the initially susceptible population will be infected during the cycle formation. This
is the size of this outbreak for this R0 parameter. In network science theory, this number quantifies
the so-called ”giant component” that indicates a phase change in the system.

Epidemic cicle control in the Beginning

R0=0.9999
10 2 R0=0.95
R0=0.9
R0=0.8
R0=0.3

10 3
r

10 4

10 5

10 6
10 6 10 5 10 4

Fig. 3. Variation of the giant component size with the initial infinitesimal fraction of
infected people in an outbreak controlled at the beginning

On the other hand, the same calculation with R0 = 0.9999999 does not have a second non-null
solution. This indicates the absence of outbreak with no formation of a large giant component. In
this case, the number of infected individuals, which is initially small, decreases instead of growing
as in the first case, and the infection is extinguished instead of propagates forming the outbreak.
To adequate the fix point model giving by equation (3.3) to predict the outbreak it is necessary to
introduce a non null ϵ ̸= 0 disturbance term at the initial value. Doing this we take into account
the nucleation of the infection which must be initiated by some infected person. This will bring
up a new fixed point solution to the fixed point equation (3.3), rϵ = CNϵ = RNϵ . If the infected
population is very small at beginning and quantitatively negligible when compared to the total of

102
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

susceptible individuals, it is then important the non-null ϵ in the fix point equation, since it gives
a quantification of the total fraction of recovered or deaths individuals even when R0 < 1. This is
the case of the absence of an outbreak of higher proportions.
Calculations for formation of the ”small giant component” are presented in Fig. 2. The log-log
graphicshows the final fraction of infected people rϵ for an initial fraction ϵ of infected individuals
during a controlled epidemic cycle.

Brazil
Reconstruction with Ridge Regression
Infected Relative Error
Removed Relative Error
0.05 Confirmed Relative Error

0.04
Relative Error

0.03

0.02

0.01

0.00
2/25/21 3/7/21 3/17/21 3/27/21 4/6/21
Days

Fig. 4. Relative errors for infected, removed and confirmed population in SIR model
solved with reconstructed parameters β and γ. The Brazilian case

Brazil

0.08 0.08

0.07 0.07

0.06 0.06
Contact Rate

0.05 0.05

0.04 0.04

0.03 0.03
1 day average 1 day average
14 days average ridge regression
0.02 0.02
2/25/21 3/7/21 3/17/21 3/27/21 4/6/21 2/25/21 3/7/21 3/17/21 3/27/21 4/6/21
Days Days

Fig. 5. In the first figure we compare β values calculated by 14-days average method
with the 1-day average estimation. In the second one the ridge regression estimation
is compared with the 1-day average estimation

103
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Brazil

1 day average 1 day average

0.10 0.10
14 days average ridge regression

0.08 0.08
Removal Rate

0.06 0.06

0.04 0.04

0.02 0.02

2/25/21 3/7/21 3/17/21 3/27/21 4/6/21 2/25/21 3/7/21 3/17/21 3/27/21 4/6/21
Days Days

Fig. 6. In the first figure we compare γ values calculated by 14-days average method
with the 1-day average estimation. In the second one the ridge regression estimation
is compared with the 1-day average estimation

Brazil
0.075 0.075

0.070 0.070

0.065 0.065
Contact Rate

0.060 0.060

0.055 0.055

0.050 14 days average 0.050 28 days average

ridge regression ridge regression

2/25/21 3/7/21 3/17/21 3/27/21 4/6/21 2/25/21 3/7/21 3/17/21 3/27/21 4/6/21
Dias Dias

Fig. 7. In the first figure we compare β values calculated by 14-days average method
are compared with ridge regression. In the second one the 28-days average
estimation is compared with ridge regression

4 Python 3.7 Implementation of COVID-19 Data

Processing
In order to automate the extraction of information, a library was developed to be capable of
extracting and processing the time series data (2.6), as well as combining these series with deterministic
solutions of the SIR equations. The result is being shown in the graphical that makes up the selection
shown in this work. It is focus on the graphs presented mainly in the data referring to Brazil.

104
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Basic Reproduction Number 0 in Brazil

3.0 3.0
0 ridge regression 0 7days avarage
2.5 0 1 day average
2.5 0 1 day average
2.0 2.0
0

1.5 1.5
1.0 1.0
0.5 0.5
2/25/21 3/7/21 3/17/21 3/27/21 4/6/21 2/25/21 3/7/21 3/17/21 3/27/21 4/6/21
3.0 3.0
0 14days avarage 0 21days avarage
2.5 0 1 day average
2.5 0 1 day average
2.0 2.0
0

1.5 1.5
1.0 1.0
0.5 0.5
2/25/21 3/7/21 3/17/21 3/27/21 4/6/21 2/25/21 3/7/21 3/17/21 3/27/21 4/6/21
Days Days

Fig. 8. Comparison of estimation of the Basic Number of Reproduction R0 for ridge

regression, 7 , 14 and 21 days average with 1 day average

The system, however, works for any country in that database and can be easily adapted to other
databases. The data made available by JHUCSSE were processed in the Python 3.7 language. A
reference on Python we quote ([15]). The Python libraries such as Pandas for extracting information
and Skit-Learning ([16]) for parameters prevision are used. In addition, other libraries such as SciPy,
Math and Matplotlib were extensively used.

5 Estimates for the Contact Rate Parameter β and the

Removal Rate Parameter γ
The main statistical tool used for estimating the parameters β and γ was the ridge regression,
which implements the Tychonoff regularization for the linear system formed with the regression of
the time series data.
The SIR equations (2.1) can be rewritten in terms of the number of confirmed cases up to a time
t, C(t) := I(t) + R(t). If we notice that

S(t) I(t) + R(t)

=1− (5.1)
N N
The second equation of the SIR model can be rewritten as:

d ln(1 − (I(t)+R(t)
)
−N N
= β(t)I(t) (5.2)
dt
while the equation for the removed people rate remains unchanged. Thus, by integrating these
equations between [t − τ, t], we obtain
∫
(I(t) + R(t)) (I(t − τ ) + R(t − τ )) t
−N ln(1 − ) + N ln(1 − )= β(t1 )I(t1 )dt1
N N t−τ

and

105
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

∫ t
R(t) − R(t − τ ) = γ(t1 )I(t1 )dt .
t−τ
τ
Defining the β and γ τ means as:
∫t
τ t−τ
β(t1 )I(t1 )dt1
β = ∫t (5.3)
t−τ
I(t1 )dt1
∫t
τ t−τ
γ(t1 )I(t1 )dt1
γ = ∫t (5.4)
t−τ
I(t1 )dt1
we have the average values for τ days with respect the total number of infected cases in these days
τ −N ln(1 − (I+R)
(t)) + N ln(1 − (I+R)
(t − τ ))
β = N
∫t N

t−τ
I(t1 )dt1
R(t) − R(t − τ )
γτ = ∫t (5.5)
t−τ
I(t1 )dt1
τ
The expression for β can be written in terms of a linear factor and a non-linear correction factor:
(I + R)(t) − (I + R)(t − τ ) −N ln(1 − N (t)) + N ln(1 − N (t − τ ))
(I+R) (I+R)
τ
β = ∫t
I(t1 )dt1 (I + R)(t) − (I + R)(t − τ )
t−τ

The linear factor can be identified as βef given by equation (5.6)

τ (I + R)(t) − (I + R)(t − τ )
β ef = ∫t (5.6)
t−τ
I(t1 )dt1
and the non-linear factor:
τ −N ln(1 − (I+R)
N
(t)) + N ln(1 − (I+R)
N
(t − τ ))
F β,nL =
(I + R)(t) − (I + R)(t − τ )
Therefore we have:
τ τ τ
β = β ef F β,nL (5.7)
Note that in the limit τ → 0 we have:
τ 1 N
F β,nL → =
1 − (I+R)
N
S
This is the inverse of the term that contributes for the system non-linearity. Therefore, the non-
linear correction factor will only be significant for a very high percentage of confirmed cases, what is
not observed in the time series data. In this way, for these first pandemic months, the contribution
of the non-linear factor is negligible and β ≃ βef .
The main statistical tool used for estimating the parameters β and γ was the ridge regression,
which implements the Tychonoff regularization for the linear system formed with the regression of
the time series data. When the regularization considers daily updating, the equations (5.6) and
(5.5) become:

C(t + 1) − C(t) 1
βL = ≃ βL (5.8)
C(t) − R(t)
R(t + 1) − R(t)
γ= ≃ γ1 (5.9)
C(t) − R(t)
By the n-days average method we mean the calculation of γ and β with equations (5.5) and (5.6)
with τ = n.

106
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Prediction of β and γ mean values using Tychonoff regularization.

b and γ
The representation for the estimated values β(t) b(t) using multi-linear regression is:

∑
J
βbJ (t) = a0 + aj β(t − j) (5.10)
j=1

∑
K
bK (t) = b0 +
γ bj γ(t − j) (5.11)
j=1

where J and K are the orders of the two filters and aj , j = 1, ..., J and bj , j = 1, ..., K are the
coefficients of the time series values that participate in the regression.

The coefficients are calculated minimizing the functional of Tychonoff

∑
T −2 ∑
J
F [a1 , ..., aJ ] := d 2 + α1
(β(t) − β(t)) a2j (5.12)
t=J j=0

∑
T −2 ∑
K
F [b1 , ..., bJ ] := d 2 + α2
(γ(t) − γ(t)) b2j (5.13)
t=K j=0

where α1 and α2 are regularization parameters.

The implementation of this optimization problem is already done in the Skit-Learn library ([16],[17]).

6 Results
The main result of this work is the interpretation of the temporal data evolution in the current
COVID-19 pandemic with data based graphs in terms of the SIR model with time dependent
parameters.
A second important result is the investigation about the Local Pandemic Cycle concept, with the
basic observation that there is a formation of a giant component even when the basic reproduction
number is less than one, R0 < 1.
Another important result is the derivation of the giant component size fix point equation (3.3)
which allow us to calculate the fraction of number of removed people at the end of an epidemic
cycle for a negligible initial disturbance. It is shown by this equation, how the parameter R0 can
control the dynamics of the epidemic cycle.
Lastly, it is shown in the work that S(t)
N
is a good simplification to the model which enable us
to compute the systems parameters by ridge regression or n-days average. It is show that these
parameters can be used to reconstruct the systems variables and arrive in values with controlled
relative error as seen in Fig. 4.

A brief summary of the graphics.

Fig. 1 shows the temporal evolution of the incomplete local pandemic cycle in Brazil. Those data
compare values from JHUCSSE data with numerical values solution of the SIR differential equation
system done with parameters estimated from ridge regression of the same data.
Fig. 2 shows the formation of a Local Pandemic Cycle in New Zealand between march and July
2020. Since the Basic Reproduction Number has been controlled to be less than one, R0 < 1, the
initial infinitesimal fraction of infect people start locally the pandemic but the pandemic die out

107
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

locally. This shown one strict correlation between the control measures adopted by local authorities
and the local control of the pandemic.
Fig. 3 shows the fact that every epidemic outbreak has the form of a giant component, even if
that component is infinitesimal when compared to the quantities of nations populations. This
type of structure appears when actions are made to reduce the Contact Rate reducing the Basic
Reproduction Number to values R0 < 1 at the beginning of the epidemic outbreak. An example of
this is shown in Fig. 2 and in Fig. 10.
Fig. 4 shows typical relative errors occurring when the dynamical systems SIR is solved with the
reconstructed parameters. We observe a good agreement between the compiled observational data
evolution and our numerical solution of the SIR model with the reconstructed parameters.
Figs. 5 and 6 present estimations for β and γ, respectively, using 14 days average and the ridge
regression method and compare it with the 1 day average estimation. The ridge regression preserves
oscillations from the data with some smoothing and this permits a more realist adequacy of the
SIR model as observed in Fig. 1.
Fig. 7 shows the estimated average Contact Rate β for 14 and 28 days, respectively, and compare
it with the reconstruction by ridge regression. We see that the ridge regression preserves the
oscillations in the observational data. An excessive smoothing is observed in these 14 and 28
averages.
Fig. 8 makes use of several averages that point to values that can be associated with the Basic
Number of Reproduction R0 using the average values for 2, 7, 14 and 28 days. Although the ridge
regressions oscilations shows the difficulty in the control of R0 , with a increase and decrease of the
parameter value, the averages shown the the parameter is in fact in a level between one and one
and half, perhaps R0 ≈ 1, 2. An important alert to authorities.

India

Confirmed Confirmed
Removed Removed
Infected Infected
12000 12000
Number of Cases using reconstructed parameters(x1000)

10000 10000
Number of Cases from JHU-CCSE data(x1000)

8000
8000

6000
6000

4000
4000

2000
2000

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 9. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The India
Data Case

108
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

China

Confirmed Confirmed
Removed 100 Removed
100 Infected Infected

80 80

Number of Cases using reconstructed parameters(x1000)

Number of Cases from JHU-CCSE data(x1000)

60
60

40
40

20
20

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 10. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The
Chinese Data Case

Fig. 9 shows the temporal evolution of the incomplete local pandemic cycle in India. Those data
compare values from JHUCSSE data with numerical values solution of the SIR differential equation
system done with parameters estimated from ridge regression of the same data.
Fig. 10 shows the temporal evolution of the incomplete local pandemic cycle in China. Those data
compare values from JHUCSSE data with numerical values solution of the SIR differential equation
system done with parameters estimated from ridge regression of the same data.

7 Conclusions
The results found are conveniently presented in a graphic form to facilitate understanding and
use it in decision making, as well as to warn about the seriousness of the consequences of the
permissiveness in acceptance of inadequate parameters values, which here means a high level of
Contact Rates.
It can be concluded that the ridge regression is an appropriated tool for estimation of parameters
for the SIR model and numerical reconstruction of observational data in short range. Besides that
Python 3.7 is an adequate computational ambient for treat this kind of data.
By inspection of the graphics, it can be concluded that the n days average can facilitate data
inspection due to its smoothing properties.
The immediate lockdown during the pandemic outbreak and social distancing are the main tools
for local pandemic control by controlling the Contact Rate β and consequently controlling the Basic
Reproduction Number R0 .
Future works involving forecast of investigated data the consequences of parameters in different
levels, adequate or not, will be developed.

109
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Acknowledgements
To Brazilian agencies CNPq and Capes for financial support.

Competing Interests
Authors have declared that no competing interests exist.

References
[1] Newman M. Networks. Oxford University Press; 2018.
[2] Kermack William Ogilvy, McKendrick Anderson G. A contribution to the mathematical theory
of epidemics. Proceedings of the Royal Society of London. Series A, Containing Papers of a
Mathematical and Physical Character. 1927;115(772):700-721.
[3] Kaplan I. Nuclear Physics. Addison Wesley; 1963.
[4] Castillo-Chavez Carlos. Mathematical models in population biology and epidemiology. New
York: Springer; 2012.
[5] Kloeden Peter E, Rasmussen Martin. Nonautonomous dynamical systems. American
Mathematical Soc; 2011.
[6] Allen Linda JS. A primer on stochastic epidemic models: Formulation, numerical simulation,
and analysis. Infectious Disease Modelling Journal; 2017.
[7] Cai Y, Kang Y, Banerjee M, Wang W. A stochastic SIRS epidemic model with infectious force
under intervention strategies. Journal of Differential Equations. 2015;259(12):7463-7502.
[8] Kudryashov Nikolay A, Chmykhov Mikhail A, Vigdorowitschr Michael. Analytical features
of the SIR model and their applications to COVID-19. Applied Mathematical Modelling.
2021;90:466-473.
[9] Prodanov Dimiter. Comments on some analytical and numerical aspects of the SIR model.
Applied Mathematical Modelling. 2021;95:236-243.
[10] Prodanov Dimiter. Analytical parameter estimation of the SIR epidemic model. Applications
to the COVID-19 pandemic. Entropy. 2021;23(1):59.
[11] Perc Matja, et al. Forecasting Covid-19. Frontiers in Physics. 2020;8:127.
[12] Novel coronavirus (COVID-19) cases data. Available:https://data.humdata.org/dataset/novel-
coronavirus-2019-ncov-cases
(Accessed from 2020 May to 2021 March)
[13] Kaipio J, Somersalo E. Statistical and computational inverse problems. Applied Mathematical
Science. Springer Science & Business Media. 2006;160.
[14] Brunton Steven L, Nathan Kutz J. Data-driven science and engineering: Machine learning,
dynamical systems, and control. Cambridge University Press; 2019.
[15] Hill C. Learning scientific programming with Python. Cambridge University Press; 2016.
[16] Raschka Sebastian. Python machine learning. Packt publishing ltd; 2015.
[17] Chen Yi-Cheng, Lu Ping-En, Chang Cheng-Shang, Liu, Tzu-Hsuan. A Time-dependent SIR
model for COVID-19 with undetectable infected people; 2020.
arXiv preprint arXiv:2003.00122

110
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Appendix
We select some countries to extract its data from the JHUCSSE spreadsheet. Results are similar
to those presented for the Brazilian Case.

Korea, South

Confirmed Confirmed
Removed Removed
Infected Infected
100 100

Number of Cases using reconstructed parameters(x1000)

80 80
Number of Cases from JHU-CCSE data(x1000)

60
60

40
40

20
20

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 11. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The South
Korea Data Case

Italy

Confirmed Confirmed
Removed Removed
Infected Infected
3500 3500

3000 3000
Number of Cases using reconstructed parameters(x1000)
Number of Cases from JHU-CCSE data(x1000)

2500
2500

2000
2000

1500
1500

1000
1000

500
500

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 12. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The Italy
Data Case

111
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Germany

Confirmed Confirmed
3000 Removed 3000 Removed
Infected Infected

2500 2500

Number of Cases using reconstructed parameters(x1000)

Number of Cases from JHU-CCSE data(x1000)

2000
2000

1500
1500

1000
1000

500
500

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 13. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The
Germany Data Case

Russia

Confirmed Confirmed
Removed Removed
Infected Infected

4000 4000
Number of Cases using reconstructed parameters(x1000)
Number of Cases from JHU-CCSE data(x1000)

3000
3000

2000
2000

1000
1000

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 14. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The Russia
Data Case

112
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Japan

Confirmed Confirmed
500 Removed 500 Removed
Infected Infected

400 400

Number of Cases using reconstructed parameters(x1000)

Number of Cases from JHU-CCSE data(x1000)

300
300

200
200

100
100

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 15. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The
Japanese Data Case

Iran

Confirmed Confirmed
2000 Removed 2000 Removed
Infected Infected

1750 1750
Number of Cases using reconstructed parameters(x1000)

1500 1500
Number of Cases from JHU-CCSE data(x1000)

1250
1250

1000
1000

750
750

500
500

250
250

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 16. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The Iran
Data Case

113
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Portugal

Confirmed Confirmed
Removed Removed
800 Infected 800 Infected

700

Number of Cases using reconstructed parameters(x1000)

600 600
Number of Cases from JHU-CCSE data(x1000)

500

400
400

300

200
200

100

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 17. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The
Portugal Data Case

Poland

Confirmed Confirmed
2500 Removed 2500 Removed
Infected Infected

2000 2000
Number of Cases using reconstructed parameters(x1000)
Number of Cases from JHU-CCSE data(x1000)

1500
1500

1000
1000

500
500

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 18. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The Poland
Data Case

114
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

Argentina

Confirmed Confirmed
2500 Removed 2500 Removed
Infected Infected

2000 2000

Number of Cases using reconstructed parameters(x1000)

Number of Cases from JHU-CCSE data(x1000)

1500
1500

1000
1000

500
500

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 19. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The
Argentina Data Case

Belgium

Confirmed Confirmed
Removed Removed
Infected Infected

800 800
Number of Cases using reconstructed parameters(x1000)
Number of Cases from JHU-CCSE data(x1000)

600
600

400
400

200
200

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 20. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The
Belgium Data Case. The recovery sheet for this country was no longer updated since
11/11/2020

115
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

US

Confirmed Confirmed
Removed Removed
30000 Infected 30000 Infected

25000
25000

Number of Cases using reconstructed parameters(x1000)

Number of Cases from JHU-CCSE data(x1000)

20000
20000

15000
15000

10000
10000

5000

5000

0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 21. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The United
State of America Data Case. The recovery sheet for this country was no longer
updated since 12/14/2020

United Kingdom

Confirmed Confirmed
Removed Removed
Infected Infected
4000 4000
Number of Cases using reconstructed parameters(x1000)
Number of Cases from JHU-CCSE data(x1000)

3000
3000

2000
2000

1000
1000

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 22. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The United
Kingdom Data Case.The recovery sheet for this country is not being updated in the
same way as other countries

116
 Roberty and Araujo; JAMCS, 36(3): 97-117, 2021; Article no.JAMCS.67121

South Africa

1600 Confirmed 1600 Confirmed

Removed Removed
Infected Infected

1400 1400

1200

Number of Cases using reconstructed parameters(x1000)

1200
Number of Cases from JHU-CCSE data(x1000)

1000
1000

800
800

600
600

400
400

200
200

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 23. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The South
Africa Data Case

Israel

Confirmed Confirmed
Removed Removed
800 Infected 800 Infected

700
Number of Cases using reconstructed parameters(x1000)

600
Number of Cases from JHU-CCSE data(x1000)

600

500

400
400

300

200
200

100

0
0
0 100 200 300 400 0 100 200 300 400
Days Days

Fig. 24. Comparing SIR model evolution states Data compiled by Johns Hopkins
University Center for Systems Science and Engineering (JHUCSSE) with Numerical
Simulations of the system (2.1) using reconstructed β and γ parameters - The Israel
Data Case

——————————————————————————————————————————————–
⃝c 2021 Roberty and Araujo; This is an Open Access article distributed under the terms of the Creative
Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits un-restricted
use, distribu-tion and reproduction in any medium, provided the original work is properly cited.

Peer-review history:
The peer review history for this paper can be accessed here (Please copy paste the total link in your browser
address bar)
http://www.sdiarticle4.com/review-history/67121

117