IEEE Computational Intelligence Magazine - November 2020 - 20

where I(t) is the number of diagnosed infections over time, t 0 is
initial time, and r is the growth rate, which can be determined
through learning from the time-series data of diagnosed infections by minimizing the mean absolute error. Cássaro and Pires
[55] utilized the time-series data of diagnosed infections collected from eight countries (Greece, Italy, Spain, Germany,
France, Netherlands, the UK, and the USA) to learn the
growth rate, r. The prediction results show that the exponential
model can accurately predict the number of confirmed cases
within 14 days when the first infection is observed; however, it
is unable to make long-term predictions.
Zhang et al. [57] proposed a logistic growth probabilistic
model that considers both the power law and the exponential
law to estimate the number of infected patients. Unlike the
exponential model that can only deal with the estimation of
uncontrolled prevalence, the logistic growth probabilistic model
is initially approximated to the exponential law, but the upper
bound of the model is set and used to reduce the growth rate.
Accordingly, the logistic growth probabilistic model is formulated as

I^t h =

N
, (4)
1 + e b - c^t - t0h

where I(t) is the number of diagnosed infections over time, N
is the predicted upper bound, b and c are the fitting coefficients that can be learned from the dataset, and t 0 is the time
when the first infection is observed. The prediction results
show that the logistic growth probabilistic model can make
long-term predictions that have a low prediction error within
3 months.
Kucharski et al. [56] designed a stochastic transmission
dynamic model to estimate the variation in transmission over
time during January and February of 2020. A dataset that consists of the COVID-19 population in or from Wuhan was collected. The transmission was modelled as a geometric random
walk. Based on the proposed stochastic transmission dynamic

FIGURE 2 A snapshot of EpiRisk.

IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE | NOVEMBER 2020

model, the probability of outbreak in other areas was estimated.
To train the proposed stochastic transmission dynamic model,
the model was fitted into four publicly available datasets: 1)
daily numbers of new global confirmed cases beginning January 26, 2020; 2) daily numbers of new confirmed cases in
Wuhan between December 1, 2019 and January 1, 2020; 3)
daily numbers of new confirmed cases in China between
December 29, 2019, and January 23, 2020; and 4) proportions
of confirmed cases on evacuation flights between January 29,
2020 and February 4, 2020.
VIII. Real-World Systems and Tools Using
Computational Intelligence for Combating COVID-19

Many industries and nonprofit organizations have been utilizing computational intelligence to develop systems or tools for
combating COVID-19. According to the report published by
the Organization for Economic Co-operation and Development (OECD) [58], these real-world systems and tools for
combating COVID-19 can be utilized to support decision
makers, the medical community, and society to manage every
stage of the COVID-19 crisis; these stages consist of detection,
prevention, response, and recovery. Based on the OECD's
report, several AI-powered tools, including BlueDot [59],
EpiRisk [60], CRUZR robot [61], Canada's COVID-19 chatbot [62] and Satellites Monitor [63], can be used for combating COVID-19. However, the details of these systems and
tools were not stated in the report or other relevant literature.
In fact, their results are still worth introducing and promoting
to the community of computational intelligence. Instead of
introducing how they work, in this section, we focus on what
they have done.
BlueDot [59] is a software that evaluates the outbreak risk
of infectious diseases caused by over 150 different pathogens,
toxins, and syndromes. In fact, COVID-19 is the most crucial
disease whose outbreak risk is detected by BlueDot. The main
technique behind BlueDot is a crawler that can scan over
100,000 official and mass media sources in 65 languages per
day. Based on the data crawled, natural language processing and
text mining are applied to extract important information for
the evaluation of outbreak risk of infectious diseases. Although
BlueDot is recognized by the OECD's report to support decision makers at the detection stage, it provides a user-centric
view that can also be utilized to calculate an individual's probability of infection (i.e., issues at the prevention stage). Unlike
BlueDot, EpiRisk [60] is a web-based application that calculates an individual's probability of infection based on a topology
structure of airline transportation networks. Since the data
source of EpiRisk is quite narrow and might miss some crucial
data, the evaluation results are doubtful, even if it is very convenient to use. For example, as shown in Fig. 2, EpiRisk shows
that the probability of infection in Taipei is higher than 45%,
which is completely untrue, with zero new confirmed domestic cases over the past 80 days in Taiwan.
When the outbreak began, how to effectively isolate infection was a key issue. The operating site might be divided into

IEEE Computational Intelligence Magazine - November 2020

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