IEEE Computational Intelligence Magazine - November 2020 - 35

I. Introduction

... in this section we propose a novel framework
he novel coronavirus (SARS-CoV-2)
has plunged the world into severe
for optimum training of an RNN using time series
disaster recently. The virus made its
analysis of the signals used in training.
way to many countries around the
globe soon after the first case was reported in
used to train a separate RNN. Similarly, all the non-stationWuhan, Hubei Province, People's Republic of China (PRC)
ary modes are used to train another RNN. The results of all
in late December [1]. As such, the World Health Organizaof the predictions are summed as the final forecast number
tion (WHO) declared the situation as a public health emerof COVID-19 cases.
gency of international concern on 30 January, 2020 [2].
India is one of the highly impacted countries, and has been
WHO officially named the disease COVID-19 when PRC
severely hit by the spread of the COVID-19 virus in many of
Center for Disease Control and Prevention (CDC) recogits states. Some researchers have sought to predict the effect of
nized the virus as a new type of coronavirus. Ever since,
lockdown measures on the spread of the virus in India and
many countries have experienced disasters due to the widehave suggested some policies to be followed by the jurisdicspreading infectious virus. This has put a huge burden on
tions to fight further spread of the virus in the country [16]-
medical centers in different countries and many different
[18]. In [16], [17], an evolutionary data analytical method called
measures have been put in place by jurisdictions to control
genetic programming was used to predict the possible impact
the spread of the virus in different countries. These measures
of COVID-19 in India. Here only two parameters, namely
are mainly in the form of lockdowns enforced in several
confirmed cases and total death count, were taken into considstages, where people are banned from congregating en messe.
eration to analyze and predict the total rise in the coming ten
Physical distancing measures can have a huge impact on the
days. The present work extends the former basic parametric
virus transmission rate [3], and in one such study the time
analysis, adding transmission rate from outbreak and the local
taken for the daily number of new cases to double was
meteorological temperature and humidity data. In this paper,
reported to increase from 2 to 4 days [4]. The optimal lockthe data from the outbreak in different parts of India have been
down policy depends on the fraction of infected and susceptaken as the case study. The source of dataset for comparison is
tible in the population. As a result, a severe lockdown
available at [19].
beginning two weeks after the outbreak was prescribed
where it can be gradually relaxed to cover 60% of the population after a month, and 20% of the population after three
II. Calculation of the Transmission Rate
months [5]. It was also recommended that the intensity of
The number of the confirmed cases has continually increased
the lockdown should depend on the gradient of the fatality
since 24 March 2020 when an outbreak was declared in differrate as a proportion of the infected, and on the assumed
ent states of India. Figure 1 shows the number of daily new
value of a statistical life [5].
confirmed cases in two of the severely affected states, namely
The effect of the Meteorological parameters on the spread
Maharashtra (Figure 1(a)) and Tamil Nadu (Figure 1(b)), since
of the COVID-19 disease has also been investigated [6], [7]. It
24 of March. These two states are taken as the case studies in
has been reported that the virus favors low temperature and
this paper.
low humidity [8]-[11]. Mortality is also shown to be affected
There have been overall, five lockdown periods in India as
by temperature and humidity variation [12]: one unit increase
of 24 March, 2020, followed by an unlock phase. The informaof temperature and absolute humidity was associated with a
tion about each lockdown phase is outlined in the Table I.
decreased COVID-19 death rate. Accordingly, temperature and
Using the number of confirmed cases corresponding to each
humidity are suggested as important factors to be considered in
lockdown phase, the value of the transmission rate in that phase
modelling of rates.
has been calculated through the following formula [20],
Some studies have focused on prediction of the number of
1 log 1 - I 1 + 1 (1)
b =
N`
`
T
I
S jj
future cases with different lockdown policies in different countries [13]-[15]. This will facilitate the investigation of the effect
where b is the mean estimated transmission rate for each lockof different measures on the future spread of the virus by
down phase, I N is the number of new infections since the
administrators and health officials.
previous lockdown, I and S represent respectively the numThis paper uses both transmission rate and meteorologiber of infected and susceptible individuals, and T is the samcal parameters (temperature and humidity) as features for
pling interval.
training a set of Recurrent Neural Networks (RNN) to
The transmission rates corresponding to the outbreak in
forecast the number of future cases of COVID-19. A sysMaharashtra and Tamil Nadu have been calculated using (1) for
tematic procedure is proposed in this paper which decomall lockdown phases of Table I. Note that the number of susposes each signal (all features as well as the signal to be
ceptible cases S in each state has been considered to be the
predicted) into its stationary and non-stationary modes. All
entire population of that state. Figures 2(a) and 2(b) show
the stationary modes that are similar in center frequency are
respectively the calculated transmission rates in Maharashtra
NOVEMBER 2020 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE

IEEE Computational Intelligence Magazine - November 2020

Table of Contents for the Digital Edition of IEEE Computational Intelligence Magazine - November 2020