J. Nig. Soc. Phys. Sci. 3 (2021) 395–405

Journal of the
Nigerian Society

of Physical
Sciences

COVID-19 Risk Factors, Economic Factors, and
Epidemiological Factors nexus on Economic Impact: Machine

Learning and Structural Equation Modelling Approaches

David Opeoluwa Oyewolaa, Emmanuel Gbenga Dadab,∗, Ndunagu Juliana Ngozic, Terang A. U.a,
Akinwumi S. A.a

aDepartment of Mathematics and Computer Science, Federal University Kashere, Gombe, Nigeria
bDepartment of Mathematical Sciences, University of Maiduguri, Maiduguri, Nigeria

cDepartment of Computer Sciences, National Open University of Nigeria, Nigeria

Abstract

Since the declaration of COVID-19 as a global pandemic, it has been transmitted to more than 200 nations of the world. The harmful impact of
the pandemic on the economy of nations is far greater than anything suffered in almost a century. The main objective of this paper is to apply
Structural Equation Modeling (SEM) and Machine Learning (ML) to determine the relationships among COVID-19 risk factors, epidemiology
factors and economic factors. Structural equation modeling is a statistical technique for calculating and evaluating the relationships of manifest
and latent variables. It explores the causal relationship between variables and at the same time taking measurement error into account. Bagging
(BAG), Boosting (BST), Support Vector Machine (SVM), Decision Tree (DT) and Random Forest (RF) Machine Learning techniques was applied
to predict the impact of COVID-19 risk factors. Data from patients who came into contact with coronavirus disease were collected from Kaggle
database between 23 January 2020 and 24 June 2020. Results indicate that COVID-19 risk factors have negative effects on epidemiology factors.
It also has negative effects on economic factors.

DOI:10.46481/jnsps.2021.173

Keywords: COVID-19, Structural Equation Modelling, Latent variables, Random forest, Boosting.

Article History :
Received: 14 March 2021
Received in revised form: 27 May 2021
Accepted for publication: 11 September 2021
Published: 29 November 2021

c©2021 Journal of the Nigerian Society of Physical Sciences. All rights reserved.
Communicated by: B. J. Falaye

1. Introduction

The negative impact of COVID-19 is felt by everyone in one
way or another. The pandemic has created a situation whereby
some people are more likely to experience severe illness be-
cause they have medical conditions that increase their risk. These
are commonly called risk factors. Examples include age, race,
gender, poverty and overcrowding, certain occupations and preg-
nancy [1]. Epidemiologic factors are definable entities that have

∗Corresponding author tel. no:

the potential to bring about a change in a health condition or
other defined outcome [2], while macroeconomic factors are a
trend or condition that comes from or applies to a broad as-
pect of an economy rather than a certain population. Common
macroeconomic factors include gross domestic product, the rate
of employment, phase of business cycle, rate of inflation and
money supply [3].
From the time when it was first discovered in Wuhan in Decem-
ber 2019, the 2019 novel coronavirus also known as COVID-
19 has quickly transmit to all regions, metropolises, and au-

395



Oyewola et al. / J. Nig. Soc. Phys. Sci. 3 (2021) 395–405 396

tonomous provinces in China and has infected many countries
in Asia, Europe, Oceanic, North America, South America, and
Africa [4]. This novel virus has posed a serious challenge to
preventing the spread of the deadly virus disease in many coun-
tries and regions and has had a great impact on economic, finan-
cial, commercial, and social development [5]. During the peak
of this pandemic, most cities in different nations of the world
have embraced closed management techniques which make busi-
nesses to operate on their own with little or no influence from
the outside world. The purpose of this is to prevent further
transmission of the virus and to lower the likelihood of more
patients being infected [6, 7]. However, in the past few months,
due to the unproductivity of most business and industrial activ-
ities in nations of the world, the great number of poor people
have been confined to their house, and this has resulted to sev-
eral social, economic and financial problems. It has also had an
enormous impact on the economic and financial development
of nations of the world [7].
COVID-19 is hypothetically a single-stranded enclosed viruses
with an DNA with size of around 26 – 32 kilobyte [8]. World
Health Organization (WHO) declared coronavirus as global pub-
lic health emergency on 30 January 2020. This is because of to
the sudden eruption of respiratory disorder. The novel coron-
avirus was classified by WHO as Severe acute respiratory syn-
drome coronavirus 2 (SARS-COV-2) and was termed the coro-
navirus disease 2019 (COVID-19) [9]. Respiratory symptoms,
fever, dry cough, fatigue, sputum production, shortness of breath,
sore throat, headache, myalgia or arthralgia, chills, nausea or
vomiting, nasal congestion, diarrhea, hemoptysis and conjunc-
tiva congestion are common symptoms of such infection. In
critical cases of COVID-19 disease, the symptoms can lead
to kidney failure, death and severe acute respiratory syndrome
[10].
In the light of the above development, the appearance of the
novel COVID-19 disease had elicited immense concerns on the
science and art of preventing disease among the populace. This
is also proved to have declining universal socioeconomic ef-
fects in due course. If the pandemic is left unchecked to con-
tinue wreaking havoc without any vigorous, reliable and sus-
tainable effort or policy to improve health of the populace, then
many economies around the globe will witness more reduced
economic activities, and many will get poorer than before [11].
The economic impact of COVID-19 on the nations of the world
cannot be overemphasized. Industrial plants and business fa-
cilities have been collapsed in a number of affected nations
[11]. Also, the delivery of goods and services through a trans-
national corporations’ worldwide network has been interrupted.
For instance, the worsening universal economic impact of the
COVID-19 endemic, and the feud between Saudi Arabia and
Russia have made Brent crude prices to sell lower $22 per bar-
rel. This happens to be the least selling price since 2003 [12].
With the impending economic downturn as a consequence of
the endemic, the situation can only be salvaged if adequate
measures are put in place [12]. For people living in underdevel-
oped and developing countries with densely-populated houses,
limited hygiene, and unavailability of funds to ease avoiding
contacts with people, the needy are at higher risk of getting in-

fected. Furthermore, the world is at the risk of seeing more
people fall below the poverty line as a result of the high cost
of medical treatment, increased economic shock, financial cri-
sis, and increased number of deaths. As these viruses transcend
borders, the global impacts will keep on spreading. It has been
reported that about 94% of businesses across the globe have
been negatively affected, and are now experiencing COVID-
19 interruptions [12]. While it is expected that the COVID-19
threat will sooner or later disappear just like as the Ebola, Zika,
and SARS viruses that have plagued the nations of the world in
the last few years. Nevertheless, social-economic impact will
still linger even after the virus is gone.
Machine learning (ML) algorithms have been applied to solve
problems in different domain by analyzing and interpreting large
quantity of data. Machine Learning has aided in the detection
and identification of diseases, as well as drug discovery, medi-
cal imaging, smart health records, radiotherapy, robotic surgery
and pharmaceutical development. Many researchers use ma-
chine learning algorithms to solve economic and financial prob-
lems.
Supervised learning, unsupervised learning, semi-supervised learn-
ing, and reinforcement learning are all forms of machine learn-
ing. How well a machine-learning system works depends on the
type of data it uses, how well the learning algorithms function.
Structural Equation Modelling (SEM) is a sophisticated multi-
variate analytic approach commonly utilized in social sciences.
It may be used for a variety of purposes, ranging from determin-
ing basic connections between variables to doing more compli-
cated studies of measurement equivalency for simpler notions.
A number of SEM analytical methods are combined. These
include comparisons of variance between and within groups,
which are usually linked to ANOVA. It also incorporates path
analysis, which involves solving equations that describe the in-
fluence of one or more variables on others in order to evaluate
the strength of their connection. As a result, path analysis il-
lustrates the predicted causal links between the variables being
investigated.
Since the beginning of the pandemic, some studies have been
conducted to provide better understanding of the COVID-19
factors that are making negative impact on the global economy.
In this paper ML algorithms and Structural Equation Modelling
(SEM) was applied to predict the impact of COVID-19 pan-
demic on global economy. ML algorithms have proved over
the years that they are very efficient and robust algorithm that
successfully cope with huge data. They can therefore be used
to prudently predict the impact of COVID-19 risk factors on
economic factors. This paper analyzes the correlation among
COVID-19 risk factors, economic factors, and epidemiology
factors and their impact on the COVID-19 crisis. Having an
understanding of this will further help in policy formulation
that will assist in mitigating against the effect of the pandemic.
Moreover, it has the potential to positively impact labor produc-
tivity and economic growth of any nation. The major contribu-
tions of this paper are:

1. A survey of different ML algorithms and SEM that have
been applied to predicting COVID-19 risk factors, eco-

396



Oyewola et al. / J. Nig. Soc. Phys. Sci. 3 (2021) 395–405 397

nomic factors, and epidemiology factors was presented.
2. Application of machine learning techniques and SEM on

epidemiological data of COVID-19 infection cases in South
Korea was discussed.

3. Performance of the all ML algorithms and SEM was eval-
uated using different performance metrics.

The rest of this paper is organized as follows: Section 2 is the
literature review. Section 3 discusses the materials and meth-
ods used in this work as well as our employed performance
measurements. The results and the discussion are presented in
section 3 and section 4 is the conclusion.

2. Literature Review

COVID-19 has captured the attention of researchers around the
world. In this section, we present reviews of the impacts of
COVID-19 on business, economies, transportation and so on.
The coronavirus outbreaks have spread to 215 countries world-
wide with a total of 43,824,534 cases, 1,165,290 deaths and
32,205,492 recovered from these diseases by 27 October 2020
[13]. The pandemic has tremendous consequences in the econ-
omy, although the exact magnitude of the effect is still un-
certain. Many countries around the world had already imple-
mented partial or absolute lockdowns [14]. Governments take
emergency steps to control the epidemic, such as social dis-
tancing, quarantine and care of the reported cases to control the
illness on one side.
The authors [15] suggested image processing of time series
crude oil price through the integration of Directed Acyclic Graph
to Convolutional Neural Network (DAG). The results indicated
that combining DAG with CNN increases forecast accuracy by
14.18%. and it was established that COVID-19 negatively im-
pacts the Nigerian crude oil price, suggesting a decreasing trend
in the crude oil prices. The effects of financial markets on the
COVID-19 pandemic was discussed in a study by [16]. Dur-
ing the period of 22 January 2020 through 17 April 2020, the
researcher uses COVID-19 confirmed cases, death and market
prices of stocks from 64 countries. The outcome was that the
stock markets have a negative reaction to growth in COVID-19
cases. In other words, stock market decreased as the number of
reported cases grew. In addition, they find proactive response
on financial markets with the rise in the numbers of suspected
cases as opposed to the increase in deaths. The developing
countries were severely impacted by the COVID-19. The au-
thors [17] mentioned the effect of COVID-19 on transportation
in Lagos, Nigeria. This study was based on an email and social
media administered to the residents of the Lagos state from 18th

to 24th May 2020, to assess the effects of COVID-19 on trans-
port in Lagos. Findings have shown that the transportation has
been negatively affected by the pandemic. There is also a posi-
tive association between COVID-19 and transportation with its
effect on people’s economic, social and religious practices.
The estimated rate of deaths in India from SARS-COV-2 for 6
weeks from day 0 to 100 on 14 March 2020 was predicted by
Ghosal et al [18] using multiple and linear regression. Findings
indicate that week 6 death counts are not statistically significant

while week 5 death count is statistically significant. For sixty-
four days, two months and three days the author [19] compiled
data from the Nigeria Center for disease control. They em-
ployed three different linear regression such as quadratic, cu-
bic, and quartic. The result shows that quartic linear regression
model with an autocorrelated error of order one performed bet-
ter. Sharif et al [20] examined the linkage within spread of
COVID-19, oil price volatility shock, the stock market, geopo-
litical risks as well as US economic policy. In their appli-
cation, they show the unprecedented effect of COVID-19 on
geopolitics, economic policy instability and the market volatil-
ity of lower rate bands, as well as oil price shocks, through
the wavelet dependent Granger causality tests. The findings
show that the COVID-19 has a considerably higher impact on
the geopolitical risk than on the US financial uncertainty.
The authors in [21] examined the effect of COVID-19 on the
correlation between crude oil and agriculture. The cross-correlation
between the crude oil of Brent and agriculture future of Lon-
don Sugar, London Wheat, USA cotton and USA orange juice
have been studied using a multifaceted cross-relationship anal-
ysis. The results demonstrated the strongest link between Brent
Crude oil and London Sugar future and three other future agri-
cultural markets. They also investigated the impact of COVID-
19 on cross-correlations between crude oil and agriculture. The
findings exhibited the greatest cross correlation between Brent
Crude Oil and the London Sugar future market among other
three future agriculture markets. The results showed that COVID-
19 persistence has become stronger and the correlation between
the multifractal relations between the crude oil and the sugar fu-
ture markets are strongest. Overall analysis show that COVID-
19 has a strong impact on the correlation between the crude oil
and selected future agriculture market and multifractal proper-
ties.
In this analysis, the authors in [22] employed four different ma-
chine learning techniques such as linear regression (LR), least
absolute shrinkage and selection operator (LASSO), support
vector machine (SVM), and exponential smoothing (ES) to pre-
dict the threatening factors of COVID-19. The study consid-
ered three types of model such as the number of newly infected
cases, the number of deaths, and the number of recoveries in
the next 10 days. The results showed that the ES performed
better than the remaining three machine techniques. Six ma-
chine learning techniques such as decision tree, support vector
machine, Naive Bayes, logistic regression, random forest, and
K-nearest neighbor algorithms were used by [23]. The model
predicted that COVID-19 patients would recover from the virus
for minimum and maximum days, that the patients with high
levels of risk would not recover from the COVID-19 pandemic,
the patients with potential recovery and those likely to recover
quickly from COVID-19. The results show that decision tree
performs better than other algorithms.
The spread of SARS appears to be influenced by the weather.
During the first 16 weeks of the pandemic, [24] conducted a
worldwide scale research that included 134,871 virologic cli-
matic demographic data from 209 nations. The relationship
between COVID 19, population density, and climate was stud-
ied using Structural Equation Modeling (SEM). The findings of

397



Oyewola et al. / J. Nig. Soc. Phys. Sci. 3 (2021) 395–405 398

the study revealed that the spread of COVID 19 is influenced
by both climate and population density. The author [25] in-
vestigated the factors that might impact public perceptions of
Indonesia’s Pembatasan Sosial Berskala Besar (PSBB). Partial
Least Squares and structural equation models were used and
data were collected from 856 respondents across Indonesia’s
provinces. The advantages of the PSBB, positive perception,
negative perception, threatening perceptions of COVID-19 and
attitude toward the PSBB policy were all utilized to evaluate
these policies. More than half of the attitudes toward PSBB
policy implementation can be explained by the model, which
takes into account perceived advantages, negative and positive
views, and the danger posed by COVID-19.
The authors in [26] used structural equation modeling to exam-
ine the relationship between socio-demographic variables (gen-
der, age, level of education, place of residence, and employment
status) and COVID-19 preventive behaviour and the threat ap-
praisal of COVID-19, fear of COVID-19, trust in COVID-19 in-
formation sources, and COVID-19 conspiracy beliefs. COVID-
19 assessment of threat, confidence in COVID-19 information
sources, and fear of COVID-19 are all significant predictors
of COVID-19 preventative activity, according to the results.
COVID-19 conspiracy theories have a negative correlation with
threat assessment and trust in COVID-19 information sources.
COVID-19 danger assessment has an important and direct role
in explaining COVID-19 phobia. A study conducted by [27]
used Structural Equation Modeling to predict how work-life
balance will be affected by factors such as their own health and
emotional well-being as well as their current relationship sta-
tus as well as their location of employment. Findings showed
that elements including physical and mental health, activities,
relationship status, and place of employment directly affected
the work-life balance. There was a notable gender gap among
dentists, with far fewer women than men. Structural equation
modeling was used by Franzen et al. [28] to simulate a group of
young individuals who were polled shortly after the conclusion
of Switzerland’s initial lockdown. To find out why and how
much they helped in averting the pandemic by following the
advice to stay at home as much as possible. They believe that
people who believe they are at danger, or who have relatives
in the risk category, are more likely to follow safety precau-
tions than people who do not believe they are at risk. Coron-
avirus social separation procedures were well followed during
the first shutdown, according to research. Young individuals
felt the virus posed little personal risk, but society as a whole
was at risk. Furthermore, the findings show that support for
preventative measures is the most significant factor in fostering
collaboration in the effort to contain the spread of COVID-19.

3. Methodology

3.1. Description of Dataset

The dataset used in this research comprises of epidemiologi-
cal data of COVID-19 infection cases in South Korea which
were obtained from Kaggle database [29] and macro-economic
data was obtained from yahoo finance [30]. The dataset is

made up of data from 23/01/2020 to 24/06/2020 recorded daily,
patient id, sex, age, country, province, city, infected by, con-
tact number, symptom onset date, confirmed date, released date
and state which consists of released, deceased and isolated.
In this study, due to the nature of the dataset, we extracted
sex, age, state, confirmed date, released date while the macro-
economic dataset obtained from yahoo finance which consists
of South Korea exchange rate (KR), Jakarta Composite Index
(JK), KOSPI Composite Index (KS) as shown in Table I. It dis-
plays sex (male=1, female=2), Age (numeric), state (released=1,
deceased=2, isolated=3), DR is obtained from subtracting re-
leased date from confirmed date. Table I shows that sex, KS,
JK and KR is negative moderately skewed while Age, state and
DR is positively skewed. The data distribution of Sex, Age,
State, JK is platykurtic since the data distribution is less than 3
while DR, KS and KR is leptokurtic since the data distribution
is greater than 3. Sex and State have a very low variance while
Age, DR, KS, JK, KR have a very high variance. This is an in-
dication that the data points are widely spaced from each other
and this may also result in high degree of error. With this in-
tuitive knowledge at hand, Table II is the variance reduction of
dataset. Fig. 1 shows a scatter plot of matrices, with bivariate
scatter plots below the diagonal, histograms on the diagonal and
Pearson correlation above the diagonal. A scatter plot shows a
relationship sex, age, state, DR, JK, KS and KR. The Pearson
correlation coefficients of the data shows a positive relationship
between sex and age, negative relationship between state and
DR, positive relationship between KS and JK, JK and KR and
KS and KR. Also, the diagonal of Fig.1 shows the histogram
of all the observed variables. There is more female than male
in the sex variable, there is more recorded cases of COVID-19
within the age of 20s than the age of 90s. In the state variable,
there is more released of patient of COVID-19 than death. DR,
JK, KS and KR are numeric values.

3.2. Structural Equation Modelling
Structural equation modeling (SEM) is also known as the causal-
ity model or covariance structural model. It is a method for
defining, estimating and evaluating the causality model [31].
It comprises a range of statistical analytical methods including
confirmatory factor analysis, variance, covariances, regression
and latent growth curve. It is a very broad and linear method of
statistical modeling that tests hypotheses according to theories.
This model was first presented by Wright [32-33]. SEM equa-
tion are split into measurement model and structural equation
model. Measurement model primarily tests the correlation be-
tween latent and significant variables while structural equation
model tests mainly causality among the latent variables. The
key characteristics of scientific research are the estimation, rel-
ativizing variables and disclosure of causality [34]. Moreover,
observable or manifest variables such as sex, age, state, DR,
JK, KS and KR can be calculated while latent variables such
as COVID-19 risk factors, epidemiological factors and macro-
economic factors cannot be directly measured as shown in Fig.
2. In such cases, regression equalities should be defined which
demonstrate how endogenous and exogenous structures are re-
lated and which benefits from a statistical technique, which has

398



Oyewola et al. / J. Nig. Soc. Phys. Sci. 3 (2021) 395–405 399

Table 1. Variance Reduction of Dataset
Variables Number

of Sam-
ple

Minimum Median Mean Variance Skewness Kurtosis

Sex 3782 0.00 2.00 1.54 0.25 -t0.19 1.05
Age 3782 0.00 40.00 40.38 408.03 0.31 2.33
State 3782 1.00 1.00 1.66 0.87 0.72 1.55
DR 3782 1.00 1.00 10.93 204.23 1.50 5.20
KS 3782 0.00 1938 1815.04 163848.07 -1.18 5.50
JK 3782 0.00 4091 3767.94 3153274.83 -0.24 1.84
KR 3782 0.00 1110 1086.97 19496.45 -3.29 29.63

Table 2. Variance Reduction of Dataset
Variables Number

of Sam-
ple

Minimum Median Mean Variance Skewness Kurtosis

Sex 3782 0.00 2.00 1.54 0.25 -0.19 1.05
Age 3782 0.00 4.00 4.04 4.08 0.31 2.33
State 3782 1.00 1.00 1.66 0.87 0.72 1.55
DR 3782 0.10 0.10 1.09 2.04 1.50 5.20
KS 3782 0.00 1.93 1.82 0.16 -1.18 5.50
JK 3782 0.00 4.09 3.76 3.15 -0.24 1.84
KR 3782 0.00 1.11 1.08 0.02 -3.29 29.63

a broad range of applications to combine measurement prin-
ciples such as SEM [35]. The risk factors of COVID-19 in-
clude age, race/ethnicity, gender, some medical conditions, use
of certain drugs, poverty and crowding, certain occupations and
pregnancy [36].
Due to the sparsity of data of COVID-19 risk factors, we con-
sidered only the sex (Sex) and age (Age) of epidemiology of
South Korea COVID-19. Epidemiology factors consists of state
(State) of COVID-19 patients and duration (DR) is obtained
from subtracting released date from confirmed date of epidemi-
ology of South Korea COVID-19. COVID-19 has catastroph-
ically affected economy. The survival of economy greatly re-
lies on the crude oil price and other macro-economic factors
[37]. In this study, we considered three macro-economic fac-
tors such as South Korea exchange rate (KR), Jakarta Compos-
ite Index (JK) and KOSPI Composite Index (KS). Fig. 3 is
the schematic diagram of structural equation modeling of im-
pact of COVID-19 risk factors on epidemiology and economic
factors. Circles are displayed as Latent variables; Squares are
displayed as Manifest, measured or observed variables; Arrows
displayed the paths from latent variables to observed variables;
Residuals and variances are indicated as double headed arrows.
The latent variables are I, Y, Z while the observed variables are
sex, age, state, DR, KR, JK, KR. The paths from latent vari-
ables to observed variable are λ1 to λ6. ϕ1 to ϕ3 double headed
curve arrows are path from each latent variables while θ1to θ10
are the double headed curve arrows for both observed and la-
tent variables. The structural equations of impact of COVID-19
risk factors on epidemiology and economic factors can be rep-

resented as:

I = α + sexβ1 + ageβ2 + ξ1 (1)

Y = ρ + stateβ3 + DRβ4 + ξ2 (2)

Z = γ + KRβ4 + JKβ5 + KS β6 + ξ3 (3)

I = Yη + Zπ + ξ4 (4)

Where I is the COVID-19 risk factors, Y is the Epidemiol-
ogy factors, Z is the economic factors, α,ρ,γ are the intercept,
β1, β2, β3, β4, β5, β6 are the predictor observable variable,
ξ1,ξ2,ξ3, ξ4 are the residual error,η,π are latent predicted vari-
able.

3.3. Machine Learning
3.3.1. Bagging (BAG)

Bagging which is an acronym for Bootstrap Aggregating is
a Parallel ensemble technique. It provides a way to decrease the
variance of prediction model throughout the training phase by
producing extra data. This is accomplished through arbitrary
sampling and substitution from the original data. Decisions
made by multiple learners can be integrated into a single predic-
tion. In the case of classification, it is clearly a vote to combine
these decisions. Models of bagging bear the same weight as
good models of bagging because an executive can use a collec-
tion of expert advice based on their previous right predictions
to achieve other outcomes. It is considered right which one gets
more votes than other groups. If more votes are expected, they
are reliable because more votes are present [38]. BAG is used
in this paper because of its capacity to minimize the variance of

399



Oyewola et al. / J. Nig. Soc. Phys. Sci. 3 (2021) 395–405 400

Table 3. Correlation Coefficient of Observed Variables
Sex Age State DR KS JK KR

Sex 1.00 0.12 -0.03 0.03 0.02 0.02 0.01
Age 0.12 1.00 0.08 -0.01 0.08 0.11 0.08
State -0.03 0.08 1.00 -0.28 0.11 -0.01 0.12
DR 0.03 -0.01 -0.28 1.00 0.20 0.25 -0.07
KS 0.02 0.08 0.11 0.20 1.00 0.76 0.15
JK 0.02 0.11 -0.01 0.25 0.76 1.00 0.23
KR 0.01 0.08 0.12 -0.07 0.15 0.23 1.00

Table 4. Summary of Impact of COVID-19 risk factor (I) on Epidemiology factor (Y ) and Economic factor (Z).
Latent Variables Estimate Std.Err Z-value p-value
I =∼sex 1.00
Age 21.84 16.62 1.32 0.19
Y =∼state 1.00
DR -68.89 199.89 -0.35 0.7
Z =∼ KS 1.00
JK 5.60 0.21 26.23 0.0
KR 0.10 0.01 13.98 0.0
Covariances
I ∼∼ Y -0.00 0.00 -0.32 0.75
I ∼∼ Z 0.00 0.00 1.31 0.19
Y ∼∼ Z -0.00 0.01 -0.34 0.73
Variance
Sex 0.24 0.01 34.89 0.00
Age 1.46 1.98 0.74 0.46
State 0.86 0.03 34.17 0.00
DR -23.53 73.83 -0.32 0.75
KS 0.07 0.00 17.30 0.00
JK 0.08 0.11 0.73 0.47
KR 0.02 0.00 43.29 0.00
I 0.01 0.00 1.27 0.20
Y 0.01 0.02 0.34 0.73
Z 0.09 0.01 19.94 0.00

a decision tree classifier. BAG enables a trade-off balance be-
tween variance and bias by reducing the variance and carefully
adjusts the prediction to an estimated result. The mathemati-
cal equation that depicts the parameters used in bagging is in
equation 5.

H
(
di, c j

)
=

M∑
m=1

αm Hm(di, c j) (5)

Where Hm is the weak classifiers, diis classified to the classes
c j and αm is the constant parameter.

3.3.2. Boosting (BST)
Boosting is a successive ensemble technique that decreases

bias error and produces outstanding prediction models. The
word ’Boosting’ describes a set of methods that transforms a
poor learner into a strong learner. Stochastic Gradient Boosting
(BST) method is a hybrid of boosting and bagging proposed by
Friedman [39]. BST is a set of learning algorithm with a combi-
nation of boosting and decision tree, which classifies the value

of all trees by weighing all trees. The new model is constructed
along the path of gradient descent of the loss function of the
previous tree. It is important to note that the loss function be-
tween classification and actual function is reduced by the train-
ing function of the classification function [40]. BST technique
was selected for use in this paper because the algorithm contin-
ues its iteration until a learner with superior results compared
to a random guess is achieved. BST approach therefore helps
in increasing the capability of machine learning and improving
prediction accuracy.
The mathematics equation of the loss function is given in equa-
tion 6 and 7:

ρ (yk, Fk (x)) =
K∑

k=0

yklog
 eFk (x)∑K

k=1 e
Fk (x)

 (6)
ŷk = −

[
∂ρ(Yk, Fk(x)
∂Fk (x)

]
= yk − Pk(x) (7)

Where y is the output variable, x is the input variables, k is the
400



Oyewola et al. / J. Nig. Soc. Phys. Sci. 3 (2021) 395–405 401

Table 5. Parameters Estimate of Structural Equation Modeling
Parameter Value
λ1 0.15
λ2 0.80
λ3 0.08
λ4 -3.54
λ5 0.77
λ6 0.99
λ7 0.23
θ1 0.98
θ2 0.36
θ3 0.96
θ4 -11.52
θ5 0.40
θ6 0.03
θ7 0.95
θ8 1.00
θ9 1.00
θ10 1.00
ϕ1 -0.01
ϕ2 -0.07
ϕ3 0.14

Table 6. Performance evaluation of Machine Learning
Algorithm RMSE MSE MASE
BAG 0.0541 0.0029 0.7758
BST 0.0512 0.0026 0.7685
SVM 0.0584 0.0034 0.8954
RF 0.0003 0.0187 0.2229
DT 0.0033 0.0577 0.8902

number of classes, Pk(x) is the probability.

3.3.3. Support Vector Machine (SVM)
SVM procedure categorizes both linear and non-linear data.

SVM uses a non-linear mapping to transform the training set to
a high level. In this new dimension, SVM explores the ideal
linear hyperplane separation as a decision limit by which the
tuples of a class of one class are split from another. There are
two class data that can be separated by a hyperplane with the
proper, non-linear upper dimensional mapping. In contrast to
the other approaches, hyperplanes are highly robust for overfit-
ting [41]. SVM is considered in this research due to its ability
to handle numerous continuous and categorical variable. SVM
technique is used in this paper because it has a low bias and
a high variance, nevertheless, the trade-off can be modified by
tuning the C parameter, which determines the number of infrac-
tions of the border permitted in training data, raising the bias
while reducing the variance. Equations 8, 9 and 10 for SVM
are stated below:

wT .x + b = 1 (8)

wT .x + b = −1 (9)

The set of inequalities can be combined to form:

y[wT .x + b] ≥ 1 (10)

The equation can be formulated as a minimization problem given
in equations 11, 12 and 13:

min
w,τ

J (w,τ) =
1
2

wT w + c
N∑

i=1

τ (11)

Subject y[wT .x + b] ≥ 1 to Lagrangian function, we then have

L (w, b,α,β) = J (w,τ)−
N∑

i=1

α
(
y
[
wT .x + b

]
− 1 + τ

)
−

N∑
i=1

βτ(12)

The optimal point of the Lagrangian function is given as

max
w,β

min
w,b,τ

L(w, b,α,β) (13)

Differentiate (12), we obtain

∂L
∂w

= 0, w =
N∑

i=1

αyx (14)

∂L
∂b

= 0,
N∑

i=1

αy (15)

401



Oyewola et al. / J. Nig. Soc. Phys. Sci. 3 (2021) 395–405 402

Figure 1. Pairwise Scatter plots of the dataset

Figure 2. Impacts of COVID-19 risk factors with respect to epidemiology and
economic factor

∂L
∂τ

= 0 (16)

The quadratic programming problem will be form by substitut-

Figure 3. Schematic diagram of Structural Equation Modeling of Impacts of
COVID-19 risk factors on epidemiology and economic factors. Circles are
displayed as Latent variables; Squares are displayed as manifest, measured
or observed variables; Arrows are displayed as paths from latent variables to
observed variables; Residuals and variances are indicated as double headed ar-
rows.

402



Oyewola et al. / J. Nig. Soc. Phys. Sci. 3 (2021) 395–405 403

ing (14), (15) and (16) to equation (12), we then have

min
α
γ (α) =

N∑
i=1

α−
1
2
αiα jyiy j K(xi, x j) (17)

Where K(xi, x j) is the kernel function, w is the weight, x is the
input, b is the bias, α,β are the positive real constants.

3.3.4. Random Forest (RF)
Random forest (RF) is a decision-making ensemble classi-

fier that has various types of trees. An arbitrary sequence of
features at each node is used to evaluate the division to create
an individual decision tree. Each tree is based on the individ-
ual values of a random variable. We are able to shape an RF
using bagging along with the selection of the random attribute,
using the CART method, in order to increase the trees. RF uses
the random linear combination of the input attributes. The sub-
cluster of features is not chosen randomly, but new attributes
are created, which reflect a linear combination of existing fea-
tures [42]. RF model will assist in the construction of numerous
decision trees and their merging to produce a more accurate and
reliable prediction. RF is an ensemble learning technique that
applies the idea of Bagging. It provides a compromise between
variance and bias by lowering the variance and judiciously fine-
tunes the prediction to a desired result.

3.3.5. Decision Tree (DT)
Decision trees (DT) are an easy model which classify by

dividing training data into pieces and mainly holding the re-
sult of each part [43]. It is a natural non-parametric super-
vised learning model, also called Classification and Regres-
sion Tree (CART) which produces accurate classifications with
easily understood regulations. DT Models transparency makes
them highly relevant for economic and financial purposes. In
addition, continuous and discrete data can be dealt with using
DT. Our choice of DT model in this work is based on it ability
to fit the training data flawlessly fine.

3.4. Performance Evaluation
Three measures such as Root Mean Square Error (RMSE), Mean
Square Error (MSE) and Mean Absolute Scaled Error (MASE)
are used to calculate the prediction efficiency of impacts of
COVID-19 risks factors with respect to epidemiology factors
and economic factors.

3.4.1. Root Mean Square Error (RMSE)
RMSE is defined as:

RMS E =

√√
1
n

n∑
n=1

(
In − În

)2
(18)

Mean Square Error (MSE)
MSE is defined as:

MS E =
1
n

n∑
n=1

(
In − În

)2
(19)

Mean Absolute Scaled Error (MASE)

MASE is defined as:

MAS E =
1
n

n∑
n=1

|In − În|
1

n−m
∑n

n=m+1 |In − În−m|
(20)

Where In is the real COVID-19 risk factors, În is the predicted
values and m is the seasonal period of In.

4. Result and Discussion

This section presents the experimental results of structural equa-
tion modeling and machine learning techniques such as Bag-
ging (BAG), Stochastic Gradient Boosting (BST), Support Vec-
tor Machine (SVM), Random Forest (RF), Decision Tree for
predicting impacts of COVID-19 risk factors with respect to
epidemiology and economic factors. We compared the per-
formances of the algorithms under consideration using Root
Mean Square Error (RMSE), Mean Square Error (MSE) and
Mean Absolute Scaled Error (MASE) to discern which is more
accurate in predicting the impacts of COVID-19 risk factors.
As stated earlier, we used data from the Kaggle database for
COVID-19 infection cases in South Korea.
Statistical correlation analysis was used to determine the strength
of the association between observed variables. Table IV is the
result of correlation coefficients of observed variables. A cor-
relation coefficient of 0.12 was noted between the sex and age
of COVID-19 risk factors, this indicates a weak positive linear
relationship between them. A correlation coefficient of -0.28
was noted between state and DR of epidemiology factors, this
indicates a weak negative linear relationship. There is a strong
positive relationship between KS and JK while between JK and
KR there is a weak positive relationship between them. How-
ever, KS and KR also show a weak positive linear relationship
of the three economic factors such as JK, KS and KR.
Table V displays the estimate of latent and observe variable,
covariance and variance of the COVID-19 risk factor (I), epi-
demiology (Y ) and economic factor (Z). The estimate, standard
error, z-value and p-value are also shown in the table. The es-
timate of latent variable I and the observed variables such as
sex and age are estimated as 21.84 with a standard error esti-
mate of 16.62, z- value is 1.52 but it was observed that there
is no statistically significant linear dependence of the mean of
I with respect to age. This means that no effect was observed.
Also, the estimate of latent variable Y and the observed variable
such as state of COVID-19 (state) and duration of COVID-19
(DR) are estimated as -68.89 with a standard error estimate of
199.89, z−value is -0.35 but it is insignificant at the 0.05 level.
There is statistically significant linear dependence of the mean
of latent variable Z with respect to JK and KR. The covariance
result of latent variable indicates that the covariance between
I, Y and Y, Z is approximately -0.00, which indicates that the
relationship is negative while the covariance result of I, Z is
approximately 0.00, which indicate that the relationship is pos-
itive. The small variance of sex, age, state, JK, KS, KR, I, Y, Z
it shows that the data points appear to be very similar to average
and to one another, while DR with high variance shows that the

403



Oyewola et al. / J. Nig. Soc. Phys. Sci. 3 (2021) 395–405 404

data points are very widespread from the average and from each
other.
Table VI is the Parameter estimate of Structural Equation Mod-
eling. There is a positive relationship between latent and ob-
served variable as shown in λ1, λ2, λ3,λ5, λ6, λ7 except in λ4
which shows a negative relationship between them. The pa-
rameter estimates of θ1, θ2,θ3,θ5,θ6,θ7, θ8,θ9,θ10 also shows
a positive relationship between them except in θ4 which shows
a negative relationship. ϕ1, ϕ2, ϕ3 is the impact of COVID-
19 risk factors (I) with respect to epidemiology factors (Y ) and
economic factors (Z). There is a negative effect of COVID-19
risk factors (I) on epidemiology factors (Y ), negative effect also
obtained on epidemiology factors (Y ) and economic factors (Z)
but positive effect was obtained from COVID-19 risk factors
(I) and economic factors (Z). Table VII report RMSE, MSE
and MASE of the extracted predicted value of COVID-19 risk
factors. RF outperform the rest of the algorithms with smaller
accuracy result than other methods, which means that the ap-
proaches are more effective than others. The findings show that
RF perform well.

5. Conclusion

Structural equation modeling (SEM) has provided a means to
understand direct impact of COVID-19 risk factors with respect
to epidemiology factors and economic factors. Latent variable
SEM has provided the necessary tools for developing several
equations to describe the COVID-19 behavioural impact frame-
work. It has the potential to quantify and test the relationships
between latent and observed variables. They measure the uni-
formity and plausibility of the assumed model in relation to
the findings observed. Furthermore, a researcher can examine
both direct and mediate relationships. Findings indicate that
COVID-19 risk factors have negative effects on epidemiology
factors. It also has negative effects on economic factors.
The result indicates that there is no statistically significant lin-
ear dependence of the mean of COVID-19 with respect to Age.
This means that no effect was observed. Also, the estimate of
latent variable epidemiology and the observed variable such as
state of COVID-19 and duration of COVID-19 is insignificant
at the 0.05 level. Also, there is a negative effect of COVID-19
risk factors on epidemiology factors, negative effect also ob-
tained on epidemiology factors and economic factors but posi-
tive effect was obtained from COVID-19 risk factors and eco-
nomic factors.
Future research areas will consider impacts of COVID-19 on
other factors such as environmental factors, socioeconomic fac-
tors, educational factors and so on.

References

[1] E.J. Williamson, A.J. Walker, K. Bhaskaran, S. Bacon, C. Bates, C.E.
Morton, H.J. Curtis, A. Mehrkar, D. Evans, P. Inglesby, J. Cockburn,
“Factors associated with COVID-19-related death using Open Safely”,
Nature, 584 (2020) 430.

[2] X. Yewwei, W. Zaisheng, L. Huipeng, M. Gifty, W. Dan, T. Weiming,
“Epidemiologic, clinical and laboratory findings of the COVID-19 in the

current pandemic: systematic review and meta-analysis”, BMC Infectious
Diseases (2020) 1.

[3] F. Zakaria, A. F. Filali, “The COVID-19: macroeconomics scenario and
role of containment in Morocco”, One Health, 10 (2020) 100152.

[4] Q. Li, X. Guan, P. Wu, X. Wang, L. Zhou, Y. Tong, R. Ren, K.S. Leung,
E.H. Lau, J.Y. Wong, X. Xing, “Early transmission dynamics in Wuhan,
China, of novel coronavirus-infected pneumonia”, N. Engl. J. Med, 382
(2020) 1199.

[5] S. Roush, H. Fast, C.E. Miner, H. Vins, L. Baldy, R. McNall, S. Kang,
V. Vund, “National Center for Immunization and Respiratory Diseases
(NCIRD) Support for Modernization of the Nationally Notifiable Dis-
eases Surveillance System (NNDSS) to Strengthen Public Health Surveil-
lance Infrastructure in the US. In 2019”, CSTE Annual Conference.
CSTE

[6] S.A. Ekanem, E.P.K. Imarenezor, C.P. Kolisah, “An Essencist Evaluation
of Socio-Economic Impacts of Coronavirus Disease (COVID-19) Pan-
demic in Nigeria”. Mediterranean Journal of Social Sciences, 11(2020),
70.

[7] A. Obioma, A.A. Reuben, A.B. Elekwachi, “Potential Impact of COVID-
19 Pandemic on the Socio-Economic Situations in Nigeria: A Huge Pub-
lic Health Risk of unprecedented Concern”, J Qual Healthcare Eco., 3
(2020) 000175.

[8] L.L. Ren, Y.M. Wang, Z.Q. Wu, Z.C. Xiang, L. Guo, T. Xu, Y.Z. Jiang,
Y. Xiong, Y.J. Li, X.W. Li, H. Li, “Identification of a novel coronavirus
causing severe pneumonia in human: a descriptive study”, Chinese med-
ical journal, 133 (2020) 1015

[9] WHO. Novel Coronavirus–China: https://www.who.int/csr/don/12-
january-2020-novel-coronavirus-china/en/. Accessed: 20 October, 2020.

[10] ational Center for Immunization and Respiratory Diseases
(NCIRD) DoVD. Coronavirus Disease 2019 (COVID-19) Sit-
uation Summary: Centers for Disease Control and Prevention.
https://www.cdc.gov/coronavirus/2019-nCoV/summary.html. Accessed:
15 November, 2020.

[11] From pandemic to poverty: Nigeria’s future with
COVID-19. (May 2020). Nairametrics. Retrieved from
https://nairametrics.com/2020/05/17/from-pandemic-to-poverty-
nigerias-future-with-COVID-19/. Accessed: 16 November, 2020.

[12] COVID-19: A Business Impact Series.
https://home.kpmg/ng/en/home/insights/2020/04/COVID-19–a-business-
impact-series.html. Accessed 16 November, 2020.

[13] Coronavirus: https://www.worldometers.info/coronavirus/ Accessed: 27,
October 2020.

[14] K.B. Ajide, R.L. Ibrahim, O.Y. Alimi, “Estimating the impacts of lock-
down on COVID-19 cases in Nigeria. Transportation Research Interdisci-
plinary Perspectives”, 7 (2020) 100217.

[15] D. O. Oyewola, A. F. Augustine, E. G. Dada, A. Ibrahim, “Predicting Im-
pact of COVID-19 on Crude Oil Price Image with Directed Acyclic Graph
Deep Convolutional Neural Network”, Journal of Robotics and Control,
2 (2020) 103-109.

[16] B.N. Ashraf, “Stock markets’ reaction to COVID-19: Cases or fatali-
ties?”, Research in International Business and Finance, 54 (2020) 101249.

[17] E. Mogaji, “Impact of COVID-19 on transportation in Lagos, Nigeria”,
Transportation Research Interdisciplinary Perspectives, 6 (2020) 100154.

[18] S. Ghosal, S. Sengupta, M. Majumder, B. Sinha, “Linear Regression
Analysis to Predict the number of deaths in India due to SARS-COV-2 at
6 weeks from day 0 to 100 cases March 14th 2020, Diabetes & Metabolic
Syndrome”, Clinical Research & Reviews, 14 (2020) 311-315.

[19] K. Ayinde, F. A. Lukman, I. Rauf, O. O. Alabi, C. E. Okon,O. E. Ayinde,
“Modeling Nigerian COVID-19 cases: A comparative analysis of models
and estimators”, Chaos Solitions and Fractals, 138 (2020) 109911.

[20] A. Sharif, C. Aloui, L. Yarovaya, “COVID-19 pandemic, oil prices, stock
market, geopolitical risk and policy uncertainty nexus in the US economy:
Fresh evidence from the wavelet-based approach”, International Review
of Financial Analysis, 70 (2020) 101496.

[21] J. Wang, W. Shao, J. Kim, “Analysis of the impact of COVID-19 on the
correlations between crude oil and agricultural futures”, Chaos, Solitons
and Fractals, 136 (2020) 109896.

[22] F. Rustam, A. A. Reshi, A. Mehmood, S. Ullah, B. W. On, W. Aslam,
G. S. Choi, “COVID-19 Future Forecasting Using Supervised Machine
Learning Models”, IEEE, 8 (2020) 101489-101499.

[23] L. J. Muhammad, M. M. Islam, S. S. Usman, S. I. Ayon, “Predictive Data

404



Oyewola et al. / J. Nig. Soc. Phys. Sci. 3 (2021) 395–405 405

Mining Models for Novel Coronavirus (COVID-19) Infected Patients’
Recovery”, SN Computer Science (2020) 1-7.

[24] A. Spad, F. A. Tucci, A. Ummarino, P. P. Ciavarella et al., “Structural
equation modeling to shed light on the controversial role of climate on
the spread of SARS-CoV-2”, Scientific Reports, 11 (2020) 8358.

[25] S.G. Purnama, D. Susanna, “Attitude to COVID-19 Prevention with
Large-Scale Social Restrictions (PSBB) in Indonesia: Partial Least
Squares Structural Equation Modeling” Front. Public Health, 8 (2020)
570394. doi: 10.3389/fpubh.2020.570394.

[26] S. Šuri, K. Martinsone, V. Perepjolkina, J. Kolesnikova, U. Vainik,
A. Ruža, J. Vrublevska, D. Smirnova, K.N. Fountoulakis, E. Ran-
cans, “Factors Related to COVID-19 Preventive Behaviors: A Structural
Equation Model”, Front. Psychol., 12 (2021) 676521. doi: 10.3389/fp-
syg.2021.676521.

[27] S. Pai, V. Patil, R. Kamath, M. Mahendra, D.K. Singhal, V. Bhat,
“Work-life balanceamongst dental professionals during the COVID-19
pandemic—A structural equation modelling approach”, PLoS ONE, 16
(2021): e0256663. https://doi.org/10.1371/journal.pone.0256663

[28] A. Franzen, F. Wohner, “Coronavirus risk perception and compli-
ance with socialdistancing measures in a sample of young adults:
Evidence from Switzerland”, PLoS ONE, 16 (2021):e0247447.
https://doi.org/10.1371/journal.pone.0247447

[29] Kaggle: https://www.kaggle.com/kimjihoo/coronavirusdataset. Ac-
cessed: 18, September, 2020.

[30] Yahoo Finance: https://finance.yahoo.com/. Accessed: 19, September,
2020.

[31] Y. Liping, C. Yuqing, P. Yuntao, W. Yishan, “Research on the evaluation
of academic journals based on structural equation modeling, Journal of
Informetrics”, 3 (2019) 304–311.

[32] S. Wright, “Correlation and Causation”, Journal of Agricultural Research
20 (1921) 557.

[33] S. Wright. S (1934), “The method of path coefficients”, Annals of Math-
ematical Statistics 5 (1934) 161.

[34] S. Kocakaya, F. Kocakaya, “A Structural Equation Modeling on Factors
of How Experienced Teachers Affects the Students Science and Mathe-
matics Achievements”, Education Research International (2014) 1.

[35] J.H. Hair, R. L. Tatham, R. E. Anderson, “Multivariate Data Analysis”,
Prentice Hall International, New York, NY, USA, 5th edition, 1998.

[36] COVID-19 Risk Factors: https://www.cdc.gov/coronavirus/2019-
ncov/COVID-data/investigations-discovery/assessing-risk-factors.html.
Accessed: 23 October, 2020.

[37] D. O. Oyewola, A. F. Augustine, E. G. Dada, A. Ibrahim, “Predicting Im-
pact of COVID-19 on Crude Oil Price Image with Directed Acyclic Graph
Deep Convolutional Neural Network”, Journal of Robotics and Control
(JRC), 2 (2020) 103-109.

[38] D. O. Oyewola, E. G. Dada, O. T. Omotehinwa, I.A. Ibrahim, “Com-
parative Analysis of Linear, Non Linear and Ensemble MachineLearning
Algorithms for Credit Worthiness of Consumers”, Computational Intelli-
gence & Wireless Sensor Networks, 1 (2019) 1-11.

[39] J. H. Friedman, “Stochastic gradient boosting”, Computational Statistics
& Data Analysis, 38 (2002) 367–378.

[40] Y. Shin, “Application of Stochastic Gradient Boosting Approach to Early
Prediction of Safety Accidents at Construction Site”, Advances in Civil
Engineering (2019) 1-9.

[41] S. Kim, J. Choi, “An SVM-based high-quality article classifier for sys-
tematic reviews”, Journal of Biomedical Informatics, 47 (2014) 153.

[42] R. Katuwal, P.N Suganthan, L. Zhang, “Heterogeneous Oblique Random
Forest”, Pattern Recognition, 99 (2019) 107078.

[43] S. Sivakumar, S. Venkataraman, R. Selvaraj, “Predictive Modeling of Stu-
dent Dropout Indicators in Educational Data Mining using Improved De-
cision Tree”, Indian Journal of Science and Technology, 9 (2016) 1.

405