ANALYZING THE EFFECTS OF DIFFERENT POLICIES AND STRICTNESS LEVELS ON MONTHLY CORONA VIRUS CASE INCREASE RATES USING MACHINE LEARNING AND BIG DATA ANALYSIS

International Journal of Data Mining & Knowledge Management Process (IJDKP), Vol.12, No.4, July 2022
DOI:10.5121/ijdkp.2022.12402 25
ANALYZING THE EFFECTS OF DIFFERENT
POLICIES AND STRICTNESS LEVELS ON MONTHLY
CORONA VIRUS CASE INCREASE RATES USING
MACHINE LEARNING AND BIG DATA ANALYSIS
Charline Chen1
and Yu Sun2
1
Irvine High School, California, USA
2
California State Polytechnic University, Pomona, USA
ABSTRACT
The corona virus is one of the most unprecedented events of recent decades. Countries struggled to identify
appropriate COVID-19 policies to prevent virus spread effectively. Although much research has been
done, little focused on policy effectiveness and their enforcement levels. As corona virus cases and death
numbers fluctuated among countries, questions of which policies are most effective in preventing corona
virus spread and how strict they should be implemented have yet to be answered. Countries are prone to
making policy and implementation errors that could cost lives. This research identified the most effective
policies and their most effective enforcement levels through data analysis of 12 common coronavirus
policies. A monthly case increase rate prediction model was developed to enable decision makers to
evaluate the effectiveness of COVID-19 policies and their enforcement levels so that they can implement
policies efficiently to save lives, time, and money.
KEYWORDS
COVID-19 Policy Effectiveness, COVID-19 Policy Enforcement Level, COVID-19 Policy Enforcement
Level and Effectiveness Consistency, Machine Learning Enabled Monthly COVID-19 Case Increase Rate
Prediction.
1. INTRODUCTION
COVID-19 was first discovered in Wuhan, China, in December 2019. A cluster of pneumonia
cases in the province was reported, and a novel coronavirus, later named COVID-19, was
eventually identified. Coronavirus diseases are a large family of viruses that appear in humans
and animals, spread through droplets and close contact, and cause mild to severe upper-
respiratory-tract illnesses [1]. The two previous coronaviruses, MERS-CoV and SARS-CoV, had
their origins from bats. Scientists linked the first coronavirus case in 2019 (COVID-19) to bats
too. Although mild symptoms of COVID-19 are common and very similar to those of the flu,
including fevers, coughs, and shortness of breath, COVID-19 is very deadly and has killed
around 6.3 million people as of June 2022 [2]. In addition, a large amount of asymptomatic
COVID-19 cases made it more difficult to track and prevent the spread of the virus. After
experiencing the MERS and SARS pandemics, many people thought countries would have been
more prepared to combat the current pandemic. However, lack of knowledge of COVID-19,
delayed notice about the virus, opened borders for international travel, and mismanagement in
handling the pandemic led to an exponential growth of cases in only a few months across the
world. Effective policy implementation with appropriate enforcement could have prevented
countries from reaching the current COVID-19 pandemic status, which has around 540 million

26
cases and 6.3 million deaths worldwide as of June 2022 [2]. Although the vaccines from some
companies (e.g. Pfizer, Moderna) have been approved by the FDA, countries will most likely still
have to live with the virus for another year or more. Analyzing and identifying the most effective
policies and their appropriate enforcement levels are critical because it is a matter of life and
death, and trillion dollar economic impacts, and will also help the world be better prepared to
handle future pandemics in an effective and timely manner before any approved vaccine is
available to the world.
Some research that has been done to identify the most effective COVID-19 policies or predict
COVID-19 cases utilize more advanced machine learning algorithms like Bayesian analysis [3]
or Poisson regression [4]. However, most research do not account for the fact that these policies
have various enforcement levels that affect the case rates. In addition, they have a narrower
research scope, with some focusing only on the U.S. or on specific policy enforcement, which
makes their research results only applicable to some countries’ COVID-19 situations.
To address the gap in current research missing policy enforcement levels impact on policy
effectiveness, this research tries to answer the question: “How effective and consistent are the
policies and their enforcement levels implemented by different countries in preventing the spread
of COVID-19?”. This research defines Effectiveness as how much a policy implementation at a
certain enforcement level can decrease the monthly case increase rate. Consistency is defined as
whether the associated monthly case increase rate will decrease accordingly as a policy
implementation strictness level increases (across all levels). This research analyzed the policy
consistency and effectiveness separately first. To measure policy effectiveness, a machine
learning model with a linear regression algorithm was developed. If the resulting coefficient of
the corresponding policy was negative, it meant that the policy helped decrease the monthly case
rate; if the coefficient was positive, it meant the policy instead led to the increase of monthly case
rates. To measure policy enforcement level consistency, Google Sheets was utilized to create
graphs for each policy and identified whether the monthly case increase rate decreased as the
policy enforcement level increased. Then the research integrated the policies’ effectiveness and
consistency analysis results to categorize the 13 policies into three overall effectiveness tiers -
most effective, partly effective, and not effective.
The machine learning model code was run 50 times and outputted the machine learning model
with the highest accuracy. The accuracy was evaluated using R-Squared, which is a
representation of how well the model fits the dataset [5]. The R-Squared of the model used in the
research was around 19%. Though it may seem relatively low, it is not bad because this research
is related to human behavior. Since human behavior is very hard to be exactly predicted, machine
learning models related to human behavior usually have lower R-Squared values. Furthermore,
the policies identified in the most effective category not only have a negative coefficient to the
policy’s enforcement level in the monthly case increase rate prediction model, but also show
consistency in their enforcement levels per the graphs.
The rest of the paper is organized as follows: Section 2 briefly discusses the challenges
encountered during research and experimentation; Section 3 illustrates the solutions to find the
answer to the research question and resolve challenges; Section 4 presents relevant details about
the experiments done. Section 5 presents related research. Finally, Section 6 gives concluding
remarks, as well as discussing future work of this project.

27
2. CHALLENGES
A list of major technical challenges encountered in this project are presented as follows:
2.1. Challenge 1: The Handling of the Incomplete Dataset Values
The data was collected from ourworldindata.org. It included policy enforcement level and
COVID-19 case numbers from 180 countries around the world from January 2020 to November
2020 (at the time of the start of this research). After downloading the data, a major observation
was that there were some null values. This could be because some underdeveloped countries did
not have the resources to collect and provide their COVID-19 case numbers, or because some
countries had not even begun to monitor COVID-19 at the time. Null values make it difficult to
develop an appropriate machine learning model. It was important for the research to address the
null value data challenge.
2.2. Challenge 2: Insufficiency of the Dataset for Calculations
In the research dataset, a column called “Monthly Case Increase Rate” was added. The formula
for this column is (Month 2’s Total Cases - Month 1’s Total Cases) / Month 1’s Total Cases. The
resulting increase rate would be placed as Month 1’s Monthly Case Increase Rate. However,
there was a problem when the data went from the last month (e.g., November 2020) of a country,
to the first month (e.g., April 2020) of the next country. In the scaled-down data set, there was no
data for December 2020 that could be used to identify November 2020’s monthly case increase
rate.
2.3. Challenge 3: Integrating Two Research Analysis Results into Conclusions
The research was split into two analysis sections: policy enforcement level graphing for
visualizing and analyzing the relationship between policy enforcement levels and monthly case
increase rates and analyzing the developed machine learning model for monthly case increase
rate predictions to identify a policy’s effectiveness. The challenge was figuring out how to
integrate results of the two analysis sections to evaluate the overall effectiveness of each policy in
a quantitative and consistent way.
3. SOLUTION
3.1. Overview of the Solution
To answer the research question: “How effective and consistent are the policies and their
enforcement levels implemented by different countries in preventing the spread of COVID-19?”,
the research established the hypothesis: If a country has a higher stringency index and enforces a
higher level of strictness for each of the 12 policies listed in ourworldindata.org, then the monthly
coronavirus case increase rate will decrease accordingly. The research takes the following
approach (as shown in Figure 1): 1) Research and Identify Data Source; 2) Review, collect, and
clean COVID-19 case data to finalize data across 168 countries in eight months for stringency
index and 12 policies and their enforcement levels from ourworldindata.org; this addressed
challenge #1 listed in section 2 by identifying the pattern of null data and removing them when
having a minimum impact on overall data analysis; 3) Integrate, calculate, and analyze the data
by using Google Sheets; this addressed challenge #2 listed in section 2 by setting the last month’s
case increase rate to that of the previous month; 4) Create graphs for stringency index and each
policy’s enforcement levels against the monthly coronavirus case increase rate to show their

28
relationships by using Google Sheets; 5) Develop monthly case increase rate prediction model
by using Google Colab and Python; and 6) Identify most effective policies and their enforcement
level through quantitative data analysis of the graphs and the coefficients of the model. This
addressed challenge #3 listed in section 2 by defining and calculating an effectiveness index
integrating the consistency between a policy’s enforcement level and monthly case decrease rate,
with the policy’s coefficient in the monthly case increase rate prediction model.
Figure 1. Overview of the Solution
3.2. Technical Details by Modules
First, ourworldindata.org was identified as the initial raw data, which has the enforcement levels
of 12 COVID-19 policies and daily COVID-19 case numbers in 2020 across around 180
countries. Later, in 2021, the website added the 13th policy, Vaccinations. This research focused
on the initial 12 major policies without the Vaccination Policy.
Second, the initial raw data across 180 countries was collected, cleaned, and integrated. When it
came to null values in the data, the research analyzed the null data pattern and found that most
null data were in the beginning three months of the dataset. This may be because some countries
were not ready to collect data when COVID-19 just started. In addition, the research found that
there were a couple of countries that did not have data in most of the months. This may be
because those few countries were not able to collect or provide COVID-19 data. Hence, the
research narrowed the timeline down to April 2020 - November 2020 because it was a period of
time where the majority of countries had already been impacted by the pandemic and were able
to collect COVID-19 policy and case data consistently. This removed most null values; the few
countries missing most data across the research time period were further removed. The remaining
small amount of null data was replaced by the monthly average of the country.
To address challenge #2 listed in section 2 - Missing Data for Calculating the Monthly Case
Increase Rate for the last month of a country’s dataset - the research took an approach of
duplicating the previous month’s monthly case increase rate for the last month. Thus, November
2020’s Monthly Case Increase Rate was set to the October 2020’s Monthly Case Increase Rate.
Though this means November 2020’s increase rate is not exact, it should only have limited
impact on the overall data analysis because it only accounts for around 12.5% of the overall data,
and the previous month’s data should be relatively close to the current month.

29
The final data set includes the country code, daily policy enforcement level for each of the 12
policies, daily COVID-19 case numbers, and calculated monthly case increase rate across 168
countries in eight months (April 2020 - November 2020).
Third, Google Sheets was used to create graphs for stringency index and each policy’s
enforcement levels against the monthly coronavirus case increase rate to show their relationships.
Fourth, Google Colab and Python were used to develop a monthly case increase rate prediction
model based on the final data set created in step 2 with the calculated “monthly case increase
rate” column as the target/label. The developed COVID-19 monthly case increase rate predictive
machine learning model (linear regression) is y = -15.816x0 + 10.607x1 - 6.335x2 + 5.184x3 -
1.853x4 - 20.557x5 - 19.102x6 + 3.194x7 + 16.658x8 - 25.029x9 - 144.590x10 - 14.722x11 +
483.090, with x0-x11 (independent variables) being the enforcement levels of each of the 12
policies(facial covering, stay at home requirements, gathering restrictions, close public
transportation, cancel public events, testing policy, international travel controls, internal
movement restrictions, contact tracing, school closures, public information campaigns, workplace
closures) respectively, and y being the predicted monthly case increase rate(dependent variable).
The model’s R-squared value is around 0.188.
Finally, the graphs (from step 3) and models (from step 4) were analyzed to draw conclusions by
integrating them into an effectiveness index. The formula to calculate the index would look at
each policy’s graph and whether the case rate decreased when the enforcement level went from
zero (no enforcement) to one, whether the case rate continued to decrease every time the
enforcement level increased, and then look at the machine learning model and check whether the
coefficient of the policy was negative or not. The policies that met all three categories would be
classified as the most effective and consistent policies; the policies that met two out of three
categories would be classified as partly effective and consistent policies; and the policies that met
one out of three categories would be classified as not effective and consistent policies. A more
detailed illustration of the approach addressing challenge #3 is described in section 4.3.
4. EXPERIMENTS/EVALUATION
4.1. Monthly Case Increase Rate vs. Policy Enforcement Level Graphs
4.1.1. Experiment Setup
After cleaning the data, Google Sheets was used to create graphs for the stringency index and
each policy’s enforcement levels against the monthly coronavirus case increase rate to show their
relationships, as shown in Figure 2 and 3 below.

30
4.1.2. Experiment Results
Figure 2. Monthly Case Increase Rate (y-axis) vs. Stringency Index (x-axis) Graph
As Figure 2 shows, when plotting the monthly case increase rate against the stringency index
(which has a value between 0 to 100 with 100 being the strictest response), it was very noticeable
that for the stringency indexes between 0-25, meaning the country was very loose on policies and
enforcement, monthly case increase rates were the highest. From then on until 100, as the
stringency index gradually became stricter, monthly case increase rates became very minimal.
The highest monthly case increase rate was 6981% at a stringency index of 11.11, while the
lowest monthly case increase rate was around 3.726% at a stringency index of 100.

31

32

33
Figure 3. Monthly Case Increase Rate(y-axis) vs. Policy Enforcement Level(x-axis) Bar Graphs for all 12
common COVID-19 policies
As all of the above policy bar graphs (in Figure 3) show, there is a consistent sharp drop in
monthly case increase rates from no policy to the first level of strictness across all policies.
Without any kind of policy in play, citizens were able to go on about their daily lives without
taking any precautions for the rapidly spreading virus. The result of that was mass virus spread
and a quick increase of case and death numbers. However, as soon as the first level of strictness
was put into play, though it was usually just a simple recommendation, it drastically lowered
monthly case increase rates because people understood more about the threat the coronavirus
posed and took action to protect themselves.
Based on the bar graph results and analysis, the ideal strictness levels for each policy to maintain
the lowest monthly case increase rate are: Facial Covering - Required outside the home at all
times; Stay at Home Requirements - Required(except essentials); Gathering Restrictions - 100-
1000 people; Close Public Transportation - Recommended closing(or reduce volume); Cancel
Public Events - Required cancellations; Testing Policy - Open public testing(including
asymptomatic); International Travel Controls - Total Border Closure; Restrictions on Internal
Movement - Recommended Movement Restriction; Contact Tracing - Comprehensive Tracing(all
cases); School Closures - Recommended; Public Information Campaigns - Coordinated
Information Campaign; Workplace Closures - Recommended.
A noticeable aspect that appeared in half of the bar graphs though, was that the strictest level of
implementation did not always mean the lowest monthly case increase rate. This was the case for
the policies: stay at home requirements, gathering restrictions, closing public transportation,
restrictions on internal movement, school closures, and workplace closures. There are a couple of

34
possible reasons that could cause this: 1) These policies may be harder to implement; 2) These
policies may take a longer time to have an impact on the monthly case increase rate; 3) These
policies may be not fully independent and partially depend on other policies’ effectiveness.
To validate reason 3), six tables were created, each of which included one of the six policies that
did not have the strictest enforcement level with the lowest monthly case increase rate. Those
tables show the average monthly case increase rate against the enforcement level of each of those
six policies. A column showing the average monthly case increase rate against the facial-covering
enforcement level to contrast the two different policies was also included.
Figure 4. Policies Whose Strictest Level not Lowest Monthly Case Increase Rate vs. Facial Coverings
All the six tables consistently showed that whenever the facial-covering policy enforcement level
average was higher, the monthly case increase rate was lower, despite the fact that the
enforcement level of each of the six policies were lower. For example, when the stay-at-home
enforcement level was at 3 the monthly case increase rate was higher than that when the stay-at-
home enforcement level was at 2. This could be because when the stay-at-home enforcement
level was at 3, the facial-covering enforcement average was around 1.65, which is lower than the
facial-covering enforcement average at around 2.21 when the stay-at-home enforcement level
was at 2. This indicates that the monthly case increase rate depends more on the mask-wearing
policy enforcement level rather than those six policies like stay-at-home requirements.
Similarly, policies like school and workplace closures also depend more on the facial covering
policy. Since some citizens really do need to be working and attending school outside of the

35
house, it is okay not to implement the strictest enforcement level as long as they strictly follow
the mask-wearing policy.
4.2. Machine Learning Model
4.2.1. Setup
To develop a monthly case increase rate prediction model and analyze each policy’s enforcement
level’s impacts on the monthly case increase rate, the research utilized Google Colab, Python
code, the sklearn library, and integrated policy related data to train and develop a Linear
Regression machine-learning-enabled prediction model (refer to Figure 5 for code). The dataset
was first split into the X dataset with predictor variables (the policy enforcement levels) and the y
dataset with the prediction variable(the monthly case increase rate). Then, these two datasets
were split into training and testing datasets with an 80%-20% split. The training X and y datasets
were used to develop the Linear Regression machine-learning-enabled prediction model. The
resulting Linear Regression model was then used for predictions on the X-test dataset, and these
predictions were compared against the y-test dataset to evaluate the model’s accuracy.
Figure 5. Machine Learning Model Code
4.2.2. Results
As the linear regression formula is y = A0x0 + A1x1 + A2x2 + A3x3 + A4x4 + A5x5 + A6x6 + A7x7 +
A8x8 + A9x9 + A10x10 + A11x11 + B with y being the predicted monthly case increase rate, A0-A11
being the coefficients, x0-x11 being the independent policy enforcement level variables for the 12
policies respectively(x0 = facial covering, x1 = stay at home requirements, x2 = gathering
restrictions, x3 = close public transportation, x4 = cancel public events, x5 = testing policy, x6 =
international travel controls, x7 = internal(domestic) movement restrictions, x8 = contact tracing,
x9 = school closures, x10 = public information campaigns, x11 = workplace closures), and B being
the intercept, the developed monthly case number increase rate predictive formula is y = -
15.816x0 + 10.607x1 - 6.335x2 + 5.184x3 - 1.853x4 - 20.557x5 - 19.102x6 + 3.194x7 + 16.658x8 -
25.029x9 - 144.590x10 - 14.722x11 + 483.090. Although most of the coefficients were negative,
meaning the specific variable contributed to lowering the monthly case increase rate, the
coefficients for the four policies, stay at home requirements, closing public transportation,
restrictions on internal movement, and contact tracing, were positive. A positive coefficient
means a higher policy enforcement level will increase the monthly case number increase rate, and
this does not align with our common sense. The possible reasons are: 1) these four policies were

36
harder to implement effectively as citizens may not follow them appropriately; 2) these four
policies may take a longer time (e.g. more than a month) to contribute to lowering the monthly
case increase rate; 3) the data used in the research was insufficient to show more accurate results.
The prediction model had an R-squared value of around 0.188. R-squared values measure the
strength of correlation in the model between the independent variable and the dependent variable
with a value between 0%-100% [5]. The model’s R-squared value was low because it predicted a
result related to/impacted by human behavior which constantly changes. R-squared values of a
prediction related to/impacted by human behavior in research is usually less than 50%. Also, the
data used to create the model may not be sufficient to create a solid and accurate machine-
learning-enabled prediction model. When collecting more data for the coming months and years,
the model can be improved. Furthermore, there could be other important independent factors
(other than the 12 policies listed in ourworldindata.org) that also have meaningful impacts on the
monthly case increase rate, but they were not included in the current prediction model.
4.3. Analysis
To address challenge #3 listed in section 2, the research defined an integrated policy effectiveness
grade integrating the analysis results on a policy from both relationship bar graphs and machine
learning enabled prediction model. As Figure 6 below shows, based on the prediction model
coefficients for each policy’s enforcement level and the enforcement level impact consistency
shown in the bar graphs Figure 3, a policy’s integrated grade can be calculated as one of 100%,
50-75%, or 25%. Policies were first given 25% if their graphs had a drop from enforcement level
0 to 1. Then, they were given 25% if they had a consistent graph, and 50% if they had a negative
coefficient. All policies that were “Most Effective and Consistent” had a grade of 100%. The
“Partly Effective and Consistent” policies that were School Closures, Workplace Closures, and
Gathering Restrictions, received a grade of 75%; the “Partly Effective and Consistent” policy
Contact Tracing received a grade of 50%. All the “Not Very Effective and Consistent Policies”
received a grade of 25%.
Figure 6. Policy Tiers Definition and Calculations
As a result, three effectiveness and consistency groups of policies were identified - 1) Most
Effective and Consistent Policies with their most effective enforcement levels (Public
Information Campaigns/Coordinated Information Campaign, Testing Policy/Open public testing
(including asymptomatic), International Travel Controls/Total Border Closures, Facial
Coverings/Required outside the home at all time, Cancel Public Events/Required cancellations);
2) Partly Effective and Consistent Policies with their most effective enforcement levels (School
Closures/Recommended, Workplace Closures/Recommended, Gathering Restrictions/100-1000

37
people, Contact Tracing/Comprehensive tracing (all cases)); 3) Not Very Effective and
Consistent Policies with their most effective enforcement levels (Restrictions on Internal
Movement/Recommended Movement Restriction, Closing Public Transportation/Recommended
losing (or reduce volume), Stay at Home Requirements/Required (except essentials)).
Based on the above analysis, my hypothesis, “If a country has a higher stringency index and
enforces the highest level of strictness for each policy, then the coronavirus case increase rate
will decrease”, is mostly (75%) correct. The stringency index against the monthly case increase
rates graph seemed to decrease exponentially as the stringency index strictness increased. This
meant that if a country had a higher stringency index, the coronavirus case increase rate would
decrease. The analysis of the monthly case increase rate against each policy’s enforcement level
showed that in six out of twelve policies, the higher level of strictness contributed to a lower
monthly case increase rate consistently across all levels. However, in the other six policies, the
highest level of strictness did not have the lowest monthly case increase rate, although monthly
case increase rates still significantly decreased from no policy to the first enforcement level.
The possible reasons for some policies not having the highest level of strictness associated with
the lowest monthly case increase rate are: 1) These policies may be harder to implement
appropriately; 2) These policies may take a longer time (e.g. longer than a month) to have an
impact on the monthly case increase rate; 3) These policies may be not fully independent and
partially depend on other policies’ effectiveness; 4) The data used in my research was
insufficient to show more accurate results. Future research should be conducted in these areas
with more data to identify more accurate relationships between monthly case increase rates and
implemented policies.
The monthly case increase rate prediction model developed based on the data from
ourworldindata.org has positive coefficients for four policies - stay at home requirements, closing
public transportation, restrictions on internal movement, and contact tracing. It does not make
sense because it seems to mean that implementing those policies will increase the coronavirus
monthly case increase rate. The possible reasons are: 1) these four policies were harder to
implement effectively as citizens may not follow them appropriately; 2) these four policies take a
longer time to contribute to lowering the monthly case increase rate; 3) the data used in my
research was insufficient to show more accurate results. Future research should be conducted in
these areas with more data.
In addition, the R-Squared value - around 0.188 - of the prediction model implied that there could
be other important independent factors (other than the 12 policies listed in ourworldindata.org)
that also have a meaningful impact on the monthly case increase rate, but are not included in the
current prediction model yet (e.g. whether a country’s leader believes in a scientific approach to
handle COVID-19, how well a country follows their government’s policy). When leaders (in a
country or a state) publicly voiced that they did not believe or were not willing to follow a
scientific approach, the case numbers seemed to increase much faster even though the overall
policy enforcement level in that area was strict in the data. Because there is a lack of information
about whether countries’ leaders believe in a scientific approach to handle COVID-19, this factor
is currently not able to be included in the model. However, this is an important area to research
more in, and an interesting factor to be taken into consideration for future research.
5. RELATED WORK
The work done by Wibbens, P. D. et al [3] utilized Bayesian analysis on big data to evaluate the
effectiveness of 11 widely implemented core COVID-19 policies in the U.S. It concluded that
though these core policies reduced growth rates for new infections, they were still not enough to

38
contain the virus. In order to bring the COVID-19 infection growth rate to near zero, higher-
impact policies, like full workplace closures, would need to be implemented. The Bayesian
approach may be useful because it updates the experiment results as more information comes in.
COVID-19 is an ongoing pandemic, making Bayesian helpful in accounting for changes as time
goes on. However, their work only focused on the U.S. and did not take into consideration
various policy enforcement levels’ (three to four levels) impact on a policy’s effectiveness. The
data collected in my research includes policy enforcement level and COVID-19 case numbers
from 168 countries around the world. Hence, my research work’s analysis is more inclusive since
it includes data from many countries across the world, and is closer to countries’ practice in the
real world because it analyzes the policy effectiveness across different policy implementation
levels.
The COVID-19 Projections model from the Institute for Health Metrics and Evaluation (IHME)
[6, 7] provides graphs of projections on total deaths, daily deaths, infections and testing, hospital
resource use, mask use, and social distancing in almost all countries. For total deaths and daily
deaths, it includes different predictions for different circumstances, such as a circumstance if
there was universal mask-wearing, a circumstance if there was a rapid vaccine rollout, and a
circumstance if mandates began easing. It also compared observed data with the predicted results.
The model, being a hybrid model that fuses elements of statistical analysis and disease
transmission model, was created to help policymakers understand how different policy decisions
can affect the trajectory of COVID-19. Based on the model they developed for the United States,
they found that universal mask-wearing could be sufficient to avoid the worst effects of epidemic
resurges. Another finding was that COVID-19-related mandates easing in states could lead to a
cumulative total of around 511,373 deaths by February 28, 2021 because relaxed mandates would
cause more open interaction and higher rates of spread. This projection model produced accurate
forecasts with the lowest median absolute percentage error being 20.2% compared to 32.6% in
many other models. Similar to my research, the IHME accounts for many different countries and
their individual pandemic-handling situations. However, the IHME model focuses more on
COVID-19 case predictions rather than also the effectiveness of COVID-19 policies. My research
focuses on both aspects as they are correlated to each other. The only mentions of COVID-19
policies in the IHME predictions are “Vaccine Coverage”, “Mask Use”, and “Social Distancing”;
but even so, there is no further detail about enforcement levels or whether they are really
effective in lowering COVID-19 case rates. My research also analyzed the effectiveness impact
of various enforcement levels (3 or 4 levels) of each of the 12 common COVID-19 policies on
the monthly case increase rate and categorized them into three effectiveness grade levels.
The work done by Arshed, N. et al [4] utilized the Panel Random Coefficient Model to estimate
the COVID-19 flattening curve and estimate the number of days it will take to reach the
flattening point. It also evaluated the effectiveness of different COVID-19 policies around the
world using Poisson regression, concluding that contact tracing, stay at home restrictions, and
international movement restrictions are most effective in controlling spread and flattening the
COVID-19 curve. However, my research took a different approach. It evaluated the effectiveness
of COVID-19 policies through a quantitative analysis approach integrating a policy’s coefficient
in a Linear Regression machine-learning-enabled monthly case increase rate prediction model,
and a policy’s implementation enforcement level’s impact on monthly case increase rates through
bar graphs.

39
6. CONCLUSION
6.1. Summary
My research proved that my hypothesis is 75% correct. It concluded that: 1) If a country has a
higher stringency index, then the monthly coronavirus case increase rate will decrease; 2) For all
of the 12 policies from ourworldindata.org(excluding vaccination), the COVID-19 monthly case
increase rate will decrease sharply (on average 89.81%) when the policies enforcement level is
increased from no measure to the first/initial level of strictness; 3) If a country enforces a higher
level of strictness for the six consistent and effective policies out of the 12 (facial covering,
cancel public events, testing policy, international travel controls, public information campaigns,
contact tracing), then the monthly coronavirus case increase rate will decrease accordingly; 4)
For the other six policies (stay at home requirements, gathering restrictions, closing public
transportation, restrictions on internal movement, school closures, and workplace closures), their
most effective enforcement level is not the strictest.
The developed COVID-19 monthly case increase rate predictive machine learning model(linear
regression) is y = -15.816x0 + 10.607x1 - 6.335x2 + 5.184x3 - 1.853x4 - 20.557x5 - 19.102x6 +
3.194x7 + 16.658x8 - 25.029x9 - 144.590x10 - 14.722x11 + 483.090, with x0-x11(independent
variables) being the enforcement levels of each of the 12 policies(facial covering, stay at home
requirements, gathering restrictions, close public transportation, cancel public events, testing
policy, international travel controls, internal movement restrictions, contact tracing, school
closures, public information campaigns, workplace closures) respectively, and y being the
predicted monthly case increase rate(dependent variable). The model’s R-squared value is around
0.188.
My research showed that the most common 12 policies provided by ourworldindata.org could be
categorized into three levels of effectiveness and consistency with a grade of 100%, 50-75%, and
25% respectively. The First Tier Policies (Most effective and consistent) consist of Public
Information Campaigns, Testing Policy, International Travel Controls, Facial Covering, and
Cancel Public Events. The Second Tier Policies (Partly effective and Consistent) consist of
School Closures, Workplace Closures, Gathering Restrictions, and Contact Tracing. The Third
Tier Policies (Not very effective and consistent) consist of Restrictions on Internal Movement,
Closing Public Transportation, and Stay at Home Requirements.
Utilizing the policy effectiveness and consistency groups, and the prediction model developed in
this research, countries and public health management organizations can identify and implement
the most effective and consistent policies at appropriate enforcement levels to lower monthly
COVID-19 case increase rates more effectively.
6.2. Current Limitations
There currently are two main limitations of the research: 1) The linear regression prediction
model’s R-Squared value is only around 0.188 which is relatively low. It may be due to the
limited data (11 months data) available at the time this research started; 2) It is also a question as
to why some policies (e.g. stay at home restrictions) enforcement levels are higher but the
monthly case increase rate does not decrease accordingly. One possible reason is that the policies
may take a longer time to have an impact, or the policies may be harder to implement
appropriately. Another reason, according to my initial research and analysis, is that the policies
may be not fully independent and partially depend on other policies’ strictness level. The research
identified that the strictest enforcement levels of the six inconsistent policies do not have the
lowest monthly case increase rate. However, the related strictest average facial covering

40
enforcement level does. The average facial covering strictness level (e.g., 2.212) was calculated
by averaging the strictness levels of facial covering when the associated policy (e.g. Stay at
Home Restrictions) was at a specific strictness level (e.g. 2) across all the countries over all
months. This meant that a lower monthly case increase rate depended more consistently on facial
covering enforcement levels rather than those six inconsistent policies’ enforcement levels
respectively.
7. FUTURE WORK
To develop a better machine learning enabled monthly case increase rate prediction model with a
higher R-Squared value, future research may include collecting more data when COVID-19
enters its second or third years, identify other potential predictors, try some other machine
learning algorithms, or develop a program to automatically run multiple times to find a best
model with the highest R-Squared value. Another interesting area to explore in future research is
to analyze and discover the most cost-efficient and effective policies in reducing COVID-19
monthly case rates through considering policy costs and integrating it with policies’ effectiveness
analysis.
REFERENCES
[1] CDC. “About COVID-19.” Centers for Disease Control and Prevention, Centers for Disease Control
and Prevention, 2020, www.cdc.gov/coronavirus/2019-ncov/cdcresponse/about-COVID-19.html.
[2] CNN. “Tracking Coronavirus' Global Spread.” CNN, Cable News Network,
www.cnn.com/interactive/2020/health/coronavirus-maps-and-cases/.
[3] Wibbens, P. D., Koo, W. W.-Y., & McGahan, A. M. (2020, December 29). Which Covid
policies are most effective? A Bayesian analysis of covid-19 by jurisdiction. PLOS ONE. Retrieved
April 10, 2022, from https://journals.plos.org/plosone/article?id=10.1371%2Fjournal.pone.0244177.
[4] Arshed, N., Meo, M. S., & Farooq, F. (2020, August 27). Empirical assessment of government
policies and flattening of the COVID19 curve. Wiley Online Library. Retrieved April 10, 2022, from
https://onlinelibrary.wiley.com/doi/10.1002/pa.2333
[5] Frost, Jim. “How To Interpret R-Squared in Regression Analysis.” Statistics By Jim, 3 Nov. 2020,
statisticsbyjim.com/regression/interpret-r-squared-regression/.
[6] “IHME: COVID-19 Projections.” Institute for Health Metrics and Evaluation, Institute for Health
Metrics and Evaluation, covid19.healthdata.org/united-states-of-america?view=total-deaths.
[7] Institute for Health Metrics and Evaluation. “COVID-19 Model FAQs.” Institute for Health Metrics
and Evaluation, Institute for Health Metrics and Evaluation, 4 Dec. 2020,
www.healthdata.org/covid/faqs.

ANALYZING THE EFFECTS OF DIFFERENT POLICIES AND STRICTNESS LEVELS ON MONTHLY CORONA VIRUS CASE INCREASE RATES USING MACHINE LEARNING AND BIG DATA ANALYSIS

More Related Content

Similar to ANALYZING THE EFFECTS OF DIFFERENT POLICIES AND STRICTNESS LEVELS ON MONTHLY CORONA VIRUS CASE INCREASE RATES USING MACHINE LEARNING AND BIG DATA ANALYSIS (20)

Recently uploaded (20)

ANALYZING THE EFFECTS OF DIFFERENT POLICIES AND STRICTNESS LEVELS ON MONTHLY CORONA VIRUS CASE INCREASE RATES USING MACHINE LEARNING AND BIG DATA ANALYSIS