IEEE Computational Intelligence Magazine - November 2020 - 17

logic models we surveyed. This may imply that we can
combine fuzzy logic and neural networks so that the
hybrid system can be applied for almost all issues relevant
to COVID-19.
2) Since the main advantage of the fuzzy logic methods is to
produce fuzzy rules that can deal with the uncertainty
from limited data, the findings from fuzzy rules are
expected to be shown in the literature. However, most relevant existing works [35]-[37] did not provide such discussions. One relevant work [34] only briefly introduced
the fuzzy rules without deep discussion. This has led to
the low applicability of these works. This is noteworthy
for future studies on fuzzy logic.
V. Evolutionary Computation for Combating COVID-19

Evolutionary computation initially creates a set of candidate
solutions and refines the set iteratively. The set of candidate
solutions at each iteration is called the population. By stochastically removing the less-desired solutions and putting small random changes in the current generation, the next generation is
produced. In biological terms, a set of solutions undergoes natural selection (or manual selection) and mutation. As a result,
the population incrementally increases in fitness. The fitness
function of the algorithm determines the goal of learning.
Evolutionary computation techniques can produce highly
optimized solutions for various problems. Many variants and
extensions have been designed for group-based risk profiling,
and they are suitable for analysis of the possible impacts of
COVID-19 and forecasting how COVID-19 will behave in
the future. Table III shows the issues addressed by existing evolutionary computation methods.
Yousefpour et al. [38] combined Susceptible, Exposed,
Infectious, and Recovered (SEIR) [43] with a multi-objective genetic algorithm that focuses on epidemic prevention
and economic concerns to estimate the early transmission
dynamics of COVID-19. Besides, Yousefpour et al. utilized
the estimation results to find the best decision rules. Two
cost functions were designed and involved in the multiobjective genetic algorithm. The first cost function represents epidemic prevention:

J 1 = / E ^ t h + A ^ t h, (1)

where E ^ t h indicates the number of exposed people at time t,
and A ^ t h indicates the number of asymptomatic infected
-people at time t. The second cost function represents economic concerns:

J 2 = - h 1 ^c 0 + c f h + h 2 q, (2)

where c 0 denotes the contact rate at the initial time, c f
denotes the minimum contact rate under the current control
strategies, and q denotes the quarantined rate of exposed individuals. Based on [38], the optimal policies were designed and
showed that treating infection control as an optimization

problem can protect countries against both disease outbreaks
and economic breakdown.
Niazkar et al. [39] adopted the multi-gene genetic programming (MGGP) to predict COVID-19 outbreaks. Since the
numbers of daily confirmed cases fluctuate, predicting a
COVID-19 outbreak is a challenging task. MGGP was originally designed for behavioral modeling, which is suitable for
modeling series with high fluctuations. The proposed method
based on MGGP showed very promising results. More specifically, the predicted number of confirmed cases of COVID-19
approximated the observations in the seven countries considered in their study. Therefore, the MGGP-based approach has
been suggested to be appropriate for the estimation of
COVID-19 outbreaks.
Salgotra et al. [40] proposed a prediction model by developing genetic programming (GP) which analyzes the possible
impact of COVID-19 in India and predicts the future behavior.
The developed GP predicted the number of confirmed cases
and numbers of death cases in the three most affected states in
India. The fitness function was designed with respect to the
mean squared error. To validate the evolved models, statistical
parameters and metrics were used to evaluate the fitness. Furthermore, the proposed GP-based models were lined with each
other by using simple linkage functions for gene size greater
than 1. The experimental results showed that the proposed GPbased models are significantly reliable for predicting the numbers of confirmed and death cases in India.
To expand the contributions of GP for predicting the possible impact of COVID-19 in India, Salgotra et al. [41] further
applied their GP to build a prediction model for forecasting the
potential effects of COVID-19 in the 15 most affected countries in the world. The prediction model estimated that the
daily confirmed cases and daily death count would result in a
negative value in China. Besides the results in China, the overall prediction results are listed in Table IV. We can find that
Brazil had the highest daily increase in the COVID-19 reproduction rate. This prediction was made at the end of May, and
Brazil's situation did fall into its worst in June. This indirectly
proves the applicability of Salgotra et al.'s study.

TABLE III Issues addressed by existing evolutionary
computation methods.
TPVP

CSVI

TRD

YOUSEFPOUR
ET AL.'S STUDY [38]



NIAZKAR ET AL.'S
STUDY [39]



SALGOTRA ET AL.'S
STUDY [40]



SALGOTRA ET AL.'S
STUDY [41]
(EXTENSION
OF [40])



DILBAG ET AL.'S
STUDY [42]

PHPM




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