Data Visualization and Analysis of COVID-19

Keshab Acharya

1. Introduction

The novel coronavirus has negatively impacted our social and academic life. The coronavirus disease, mainly referred to as Covid-19, has taken the lives of many people. In the rise of covid cases, the United States was impacted by them more than any country in the world. The covid cases started rising around late January and increased exponentially throughout the months. This immediate rise in the cases has surprised everyone. If we closely observe the impact of covid in all the countries in the world, it clearly shows that United States was the one that had the highest impact.

This project is a simple analysis of covid cases throughout the world. We will first start visualizing data just for United States and we will explore more data through different datasets. The project has been divided into multiple parts, each with their own purpose. This project largely consists of visualization and analysis as this is the best way to see the impact of covid around the world. It would make most sense to visualize how many cases and deaths have happened throughout the world using the datasets. Therefore, the majority of this project consists of plots.

In addition to visualization, we will try to analyze the data with predictions. One of the fundamentals of data science is to predict values based on models. We will map a model based on our data and try to predict cases for next couple of days. We will also do some visualizations using maps which give interesting results and a different way to interpret our data.

2. Preparing the Data for U.S

First, we will use a dataset from Kaggle that has a list of counties with their respective state, the count of covid cases, along with their deaths. We can delete all the entries with null values since they won't contribute to the plot. The dataset is updated every day since we are still in a pandemic and the cases are still rising every day. Therefore, the data analysis might not perfectly match the true increase in cases, but it should be an accurate analysis as of this writing.

The dataset for the counties can be accessed in Kaggle through US Counties COVID-19 Dataset

We will first start by importing some basic libraries we will need to start visualizing the data. We will import more libraries later as we do more analysis. To install these libraries we can simply do pip install [library name]

# Starting libraries we will use 
import pandas as pd               # Pandas for accessing dataset
import numpy as np                # Numpy for manipulating numbers and calculations
import matplotlib.pyplot as plt   # MatplotLib for graphs
import seaborn as sns             # Seaborn for beautifying graphs 

Now that we have some libraries ready to use, we will access the dataset using pandas and we will look at a small sample of the dataset.

data = pd.read_csv("covid-us-counties.csv")
data.sample(10)

	date	county	state	fips	cases	deaths
442509	2020-08-17	Washoe	Nevada	32031.0	6297	127.0
764733	2020-11-24	Washakie	Wyoming	56043.0	338	7.0
470814	2020-08-26	Carter	Kentucky	21043.0	113	3.0
432401	2020-08-14	Clay	Minnesota	27027.0	795	40.0
336811	2020-07-15	Creek	Oklahoma	40037.0	231	9.0
248738	2020-06-17	Buffalo	South Dakota	46017.0	55	0.0
155704	2020-05-18	Lafayette	Missouri	29107.0	68	1.0
136132	2020-05-11	Wyoming	West Virginia	54109.0	1	0.0
812527	2020-12-09	Guanica	Puerto Rico	72055.0	213	NaN
124556	2020-05-08	New Haven	Connecticut	9009.0	8887	669.0

As we can see, we have a NaN value for Puerto Rico, which is indeed a territory of the US. The NaN values are a replacement for no data reported in that county or state. This will hamper us from putting a graph for that row. Therefore, it will be better if we remove those entries right now and then start the visualizations. We will remove all the rows that have NaN values using pandas. After removing the Nan values, there should be valid data for every row and column in this dataset.

There is also a column named fips that has a specific code for counties which may vary for metro areas. Since we are visualizing the rise in cases, we do not have great use of fips. Hence, we will first remove the fips column from our database and then take care of the NaN values accordingly.

# Remove column name 'fips' 
data = data.drop(['fips'], axis = 1)
# Remove all the rows that have NaN values in them
data.dropna()
# Make sure the date is in date format
data['date'] = pd.to_datetime(data['date'], errors='coerce')
data_copy = data # For later reference

data_copy.sample(5)

	date	county	state	cases	deaths
620744	2020-10-11	LaMoure	North Dakota	124	0.0
276879	2020-06-26	Horry	South Carolina	2582	42.0
740780	2020-11-17	Golden Valley	North Dakota	139	0.0
12657	2020-03-26	Floyd	Georgia	17	1.0
428977	2020-08-13	Iberia	Louisiana	2588	74.0

3. Visualizing US data through plots

The table has all the information we need to start plotting. It is important to note that the cases accumulate through time. This means the cases do not represent new cases, but rather represent total cases at that date.

We will start by doing simple plots for each month through all the counties. The plot will be of cases and the corresponding deaths. We will divide the dataset, group them by date and then plot each count of cases reported in that day. The plot will have points scattered everywhere as the data size is huge and that we are compressing most of the dataset into segments by months.

Before the plots, we can make predictions that the points will be more clustered as the month increases. January should have the lowest number of points(or cases) plotted since we had very less cases that month. On contrary, November and December should have the most about of points in the plots since reportings of cases were highest throughout those months.

# Making a copy of the data so we can plot
month_table = data_copy[['date', 'cases', 'deaths']].copy()
#month_table.groupby(month_table['date'].dt.strftime('%B'))[['cases', 'deaths']].apply(sum)
plt.style.use('seaborn')
plt.rc('figure', figsize=(20,8))

# Group data by each month and graph a plot for each month
for month, df in month_table.groupby(month_table['date'].dt.strftime('%B'), sort=False):        
    x = df['date']
    y = df['cases']
    plt.scatter(x, y, cmap = 'Spectral', edgecolor='k', alpha=1, c=df['deaths'])
    plt.xlabel('Date', fontsize=12, fontweight='bold')
    plt.ylabel('Cases', fontsize=12, fontweight='bold')
    plt.title(month, fontsize=14, fontweight='bold')
    plt.show()

Looking at the plots, there seems to be a huge difference in number of cases for each month. Our prediction correctly matches the plots. There are interesting points in the plots and we can also observe a lot of inliers and outliers in the plots. The plots depict how quickly the cases increased as the months pass. This is evident from the fact that the points are getting more as more clustered as the months pass.

In January, the cases were consistently 1 with 1-2 outliers of 2 cases. In February, we start to see some increase in cases where the cases are passing 10. March is when the number of cases increase exponentially. We see that the cases are in thousands. Even though the data is skewed more towards the bottom, there is still some points that are rising exponentially in the plot. This means that at least one counties had an exponential increase in cases in that month. Most counties had cases that are less than 10000 at that time. We can see similarity in the points for the next few months, except with more and more data scattered in the plots as well as increase in the size of the y-axis which represents the number of cases. If we jump to December, we can clearly see that there is bigger split in datapoints. Even though most points are skewed near the bottom, there are points increasing at the top. Most counties may have had cases up to 100 thousands by December, however, some counties had cases more than 400 thousands. These counties maybe are the ones from New York and New Jersey.

The average cases reported in each state depicts the correct increase in cases with few expections of wrong data. As we can see in the plot, D.C. seems to have highest average number of cases which doesn't correctly depict how many cases there are in the state. This might be because of the fact that D.C barely had any cases until late 2020. So the average increase is very high for that region. The average increase for New York and New Jersey are also very high.

4. Data Analysis of top impacted countries

In this section, we will extend our analysis to not just include the US, but the entire world. We will start with top impacted countries. The dataset includes in each row a country, the number of cases in that country, date, deaths in that country, as well as other information about the country. In our case, we will mostly focus on the number of cases and the deaths for each country.

# Getting covid data for the entire world 
world_data = pd.read_csv("covid-data.csv")
world_data.sample(5)

	iso_code	continent	location	date	total_cases	new_cases	new_cases_smoothed	total_deaths	new_deaths	new_deaths_smoothed	...	gdp_per_capita	extreme_poverty	cardiovasc_death_rate	diabetes_prevalence	female_smokers	male_smokers	handwashing_facilities	hospital_beds_per_thousand	life_expectancy	human_development_index
25029	JPN	Asia	Japan	2020-06-30	18615.0	139.0	105.143	972.0	0.0	1.000	...	39002.223	NaN	79.370	5.72	11.2	33.7	NaN	13.05	84.63	0.909
47425	CHE	Europe	Switzerland	2020-03-05	114.0	24.0	15.143	1.0	NaN	0.000	...	57410.166	NaN	99.739	5.59	22.6	28.9	NaN	4.53	83.78	0.944
29330	LUX	Europe	Luxembourg	2020-08-27	7928.0	0.0	41.571	124.0	0.0	0.000	...	94277.965	0.2	128.275	4.42	20.9	26.0	NaN	4.51	82.25	0.904
43170	SYC	Africa	Seychelles	2020-03-22	7.0	0.0	0.714	NaN	NaN	0.000	...	26382.287	1.1	242.648	10.55	7.1	35.7	NaN	3.60	73.40	0.797
17591	FRA	Europe	France	2020-07-27	222508.0	2488.0	917.000	30214.0	18.0	4.571	...	38605.671	NaN	86.060	4.77	30.1	35.6	NaN	5.98	82.66	0.901

5 rows × 52 columns

As we can see, the dataset contains 52 columns. These columns include a lot of information about the country. However, our focus will be on the cases and the deaths for each country. We have many NaN values which we need to take care of as well as the date which might not be in the datetime format.

We will first do analysis of countries will the most cases. I have choose the United States, India and Italy for analysis as these are the countries will enourmous number of cases. We have already analyzed cases in the United States by looking at all the counties and states, however, in this section we are doing a broad analysis on a country level. This will help us to compare the cases in the United States with the cases in other countries like India and Italy. All three countries were significantly impacted by Covid, therefore, it helps to visualize the increase in cases and the deaths in these countries.

We will create a subset of the entire dataset which only includes the countries United States, India and Italy. This dataset still consists of every date of reporting. In other words, we have entry for the cases for every date of reporting. The date starts as early as January and ends on December 17th. The dataset updates everyday, therefore, this dataset may not reflect every date these countries had covid since we are still in a pandemic and the cases are still increasing. At the time of writing this project, the current date is reflected in the dataset which is december 17th, 2020. There will also be a new column named mortality rate, which calculates the mortality rate for each date.

# Convert date to date format
import datetime as dt
world_data['date'] = [dt.datetime.strptime(x, '%Y-%m-%d') for x in world_data['date']]

import warnings
warnings.filterwarnings("ignore")
# Lets look at United States data
# We will create a subset from the database
countries = ['United States', 'India', 'Italy']
country_data = world_data[world_data.location.isin(countries)]

# Add a new column for the mortality rate
country_data['mortality_rate'] = country_data.apply(lambda row: row['total_deaths']/
                                                    row['total_cases'], axis=1).fillna(0)

# Make date the index
country_data.set_index('date', inplace=True)
country_data[['location', 'new_cases', 'total_cases', 
              'new_deaths', 'total_deaths', 'mortality_rate']].tail()

	location	new_cases	total_cases	new_deaths	total_deaths	mortality_rate
date
2020-12-13	United States	191142.0	16334361.0	1389.0	299293.0	0.018323
2020-12-14	United States	192846.0	16527207.0	1484.0	300777.0	0.018199
2020-12-15	United States	198766.0	16725973.0	2984.0	303761.0	0.018161
2020-12-16	United States	247403.0	16973376.0	3668.0	307429.0	0.018112
2020-12-17	United States	236211.0	17209587.0	3345.0	310774.0	0.018058

Now that the dataset only contains the three countries [United States, India, Italy], we can start doing our plots. We will do four plots which will be for:

• New cases
• New deaths
• Total cases
• Total deaths

plt.style.use('seaborn')

# Graph cases and deaths in new graphs
fig, axes = plt.subplots(nrows=2, ncols=2, figsize=(16,16))

# Plot each graph
country_data.groupby('location')['new_cases'].plot(ax=axes[0,0], legend=True)
country_data.groupby('location')['new_deaths'].plot(ax=axes[0,1], legend=True)
country_data.groupby('location')['total_cases'].plot(ax=axes[1,0], legend=True)
country_data.groupby('location')['total_deaths'].plot(ax=axes[1,1], legend=True)

# Set and label the axis for each plot
axes[0, 0].set_title('New Cases', fontweight='bold')
axes[0, 0].set_xlabel('Date', fontweight ='bold')
axes[0, 0].set_ylabel('Cases', fontweight ='bold')

axes[0, 1].set_title('New Deaths', fontweight='bold')
axes[0, 1].set_xlabel('Date', fontweight ='bold')
axes[0, 1].set_ylabel('Deaths', fontweight ='bold')

axes[1, 0].set_title('Total Cases', fontweight='bold')
axes[1, 0].set_xlabel('Date', fontweight ='bold')
axes[1, 0].set_ylabel('Cases in millions', fontweight ='bold')

axes[1, 1].set_title('Total Deaths', fontweight='bold')
axes[1, 1].set_xlabel('Date', fontweight ='bold')
axes[1, 1].set_ylabel('Deaths', fontweight ='bold')

Text(0, 0.5, 'Deaths')

The four plots show the rise in cases and deaths for each month for all three countries. There seem to be various results for each plot. We can analyze each one separately.

New cases plot

• United States seems to have the highest increase in new cases whereas India and Italy have cases that are slowly decreasing in December.
• Italy was ahead of the US in mid-September.

Total cases plot

• This plot closely mimics the plot for new cases. The United States still predominates the increase in cases and in December, the cases are 1.75 million. There seems to be an exponential increase.

New deaths plot

• The lines here are swinging back and forth for the U.S. Regardless, the new deaths are still highest for the US. The new deaths are as high as 3500 which represents the deaths reported each day. This is a terrible situation.

Total deaths plot

• Just like the other plots, the US is the highest in deaths with over 300 thousand deaths in December. As of now, the rate of deaths is horrible.

4.1 Data Analysis using barplots

We will now use bar plots to visualize cases in the top 10 countries. As of now, there are hardly any countries that do not have any cases of coronavirus. This would be overwhelming to plot, therefore, we will pick the top 10 countries that have the highest amount of cases of coronavirus and graph them using bar plots.

We will be using the same database and also the same information to do the plots. The dataset contains data for each country as well as the world overall, therefore, we need to disregard that. We will also need to handle duplicates as there may be duplicates for a lot of countries. We will use a function this time to do the plotting so that we don't have to use the same code to do the same plot for cases and deaths.

world_copy = world_data.loc[~(world_data['location'].isin(['World']))]

# Group by location and date
world_copy = pd.DataFrame(world_copy.groupby(['location', 'date'])
                          ['location', 'total_cases', 'total_deaths'].sum()).reset_index()

world_copy = world_copy.sort_values(by = ['location', 'date'], ascending=False)
# Get rid of any duplicates in the dataset
world_copy = world_copy.drop_duplicates(subset = ['location'], keep='first')
# Lets do bar plots of top 10 countries
world_copy.reset_index(drop=True)

# Make a function to plot cases and death vs time
def make_plot(df, group, column, title, x_axis, y_axis):
    fig, ax = plt.subplots(1, 1, figsize=(20, 8))
    pal = sns.color_palette("tab10")
    df = df.sort_values([column], ascending=False).reset_index(drop=True)
    plot = sns.barplot(df[group][0:10], df[column][0:10], palette=pal)
    plot.set_title(("Plot for top 10 {} - ").format(title), fontweight='bold', fontsize=14)
    plot.set_xlabel(x_axis, fontweight='bold', fontsize=12)
    plot.set_ylabel(y_axis, fontweight='bold', fontsize=12)
    
    plt.show()
    
# Make bar plots for total cases and total deaths
make_plot(df=world_copy, group='location', column='total_cases', 
          title='countries with highest cases', 
          x_axis='Countries', y_axis='Total cases in millions')

make_plot(df=world_copy, group='location', column='total_deaths', 
          title='countries with highest deaths', 
          x_axis='Countries', y_axis='Total deaths')

As expected, the United States is a clear winner for both the cases and deaths. After the United States, we see a mismatch of countries for each plot. As for the cases, India seems to be the second-highest. However, for deaths, Brazil seems to be the second-highest. Brazil and India switch around the second and third positions in both plots. It is surprising to see that Italy and Spain follow 4-5 countries in both plots even though they had a rapid increase in mid-2020.

4.2 Predicting future covid cases for US

Data Science is famous for predicting data using models. Python provides many libraries for fitting models. Mainly, we need to separate the dataset into two parts, one for training and one for testing. When we are testing the model, we are trying to see how accurate the model fits the data and also see predicted values using the same model. Although are also many ways to fit a model, I will use Scipy which has a function scipy.optimize.curve_fit which fits a curve to the model.

Since the dataset is enormous, we will only use United States to fit a model. The predictions will be the cases for the United States only and not the entire world. This makes our model much easier and smoother.

We will start by creating a function that will take three parameters:

• df: the dataframe we are using to create the model
• title: title to give for the plot
• delta: how many days we are chopping off the data so that we don't include in for training. Instead, we will use it for validation.

We will start will delta as 10 which means we will use data until 10 days before and then compare the actual cases from 10 days before to the predicted one to see how much they match. We will then use the model to predict the cases for 10 more days. We will use some datapoints to create the model, and then we will use the rest for validation. We will use exponential function to fit our curve. There are many ways to fit the curve, however, if we look at the cases for United States, we can see that there is an exponential increase in the cases. The model will be more accurate since the cases are increasing exponentially (which we saw in our previous plots). It will make more sense to explain the coefficients and the predictions after we see the plot and the values.

import numpy as np
import scipy

# Function parameters:
# df - the dataframe
# title - the title of teh plot
# Delta: how many days we are chopping off for training
def exponential_data_model(df, title, delta):
    df = df.sort_values(by=['date'], ascending=True)
    df['x'] = np.arange(len(df)) + 1 # Column x in dataframe 
    df['y'] = df['total_cases'] # column y in the dataframe
    
    # Chop off data for validation and testing
    x = df['x'][:-delta] # Remove delta number of data points (so we can predict)
    y = df['y'][:-delta] 
    
    # Use exponential function to fit the curve
    fit_model = scipy.optimize.curve_fit(lambda t, a, b: a*np.exp(b*t), x, y, p0=(100000, 0.1))
    
    # Extract coefficients
    A, B = fit_model[0] # Coefficients
    print("Coefficients =====")
    print("value of A: ", A)
    print("value of B: ", B)
    print("=======================")
    
    x = range(1, df.shape[0] + 1)
    # Fit the model for predictions
    y_fit = A*np.exp(B*x)
    
    # Print the fit model
    fig, ax = plt.subplots(1, 1, figsize=(20, 8))
    g = sns.scatterplot(x=df['x'][:-delta], y=df['y'][:-delta],
                   label='Confirmed cases from the model creation', color='green')
    g = sns.scatterplot(x=df['x'][-delta:], y=df['y'][-delta:],
                   label='Confirmed cases for model validation', color='red')
    g = sns.lineplot(x=x, y=y_fit, label='Predicted Values', color='blue')
    
    # Our dataset includes 330 dates of data 
    # We will create our model using those 330 datapoints 
    # and predict the next 10 based on that model
    x_future = range(330, 340)
    y_future = A*np.exp(B*x_future)
    print("Expected cases for the next 10 days: \n", y_future)         
    plt.xlabel('Days since first case', fontweight='bold', fontsize=14)
    plt.ylabel('Total cases', fontweight='bold', fontsize=14)
    plt.title('Confirmed cases along with projected cases for ' + title, 
              fontweight='bold', fontsize=16)
    plt.xticks(rotation=90)
    ax.grid(color='yellow', linestyle='dotted', linewidth=0.75)
    plt.show()

# Use the world dataset to fit the model and print the plot
world_df = world_data.loc[~(world_data['location'].isin(['World']))]
world_df = pd.DataFrame(world_df.groupby(['location', 'date'])
                          ['total_cases', 'total_deaths'].sum()).reset_index()
world_df = world_df.sort_values(by = ['location', 'date'], ascending=False)

# Only include data for United States
us_data = world_df[world_df['location']=='United States']
us_df = us_data.copy()
exponential_data_model(us_df, "United States", 10)

Coefficients =====
value of A:  494786.69351850136
value of B:  0.010526662169295533
=======================
Expected cases for the next 10 days: 
 [15961346.95912357 16130254.12092161 16300948.70262724 16473449.61918694
 16647775.98571012 16823947.11958729 17001982.54263067 17181901.98323742
 17363725.37857573 17547472.87679419]

Fit Results

A and B represent our coefficients which define the curve for the model. For the predictions, we can see that the predictions for the next 10 days start from 15.9 million. At the time of this writing, December 17th, the cases are closer to that number. The model predicts that in 10 days the cases will be as high as 17.5 million for the United States. In the plot, we can see three lines-

• Blue: Our predicted curve which is exponential
• Green: the curve that reflects actual cases (our model)
• Red: The cases used for validation

Note

The predicted curve increases exponentially, but it is slightly lower than the confirmed cases. Since its slightly lower, the predicted values might be also slightly lower than the actual value. For example, the predicted cases in the next 10 days might be off by a million since the curve is little lower than the actual curve of the model.

5. Data Analysis of the Whole World

We will continue our visualization and analysis but we will use different datasets. I am doing the same visualizations to find consistency in data. In other words, these datasets might be more consistent and accurate than the others. We will also use maps to visualize our data since maps are very useful and effective in communicating the data.

Plotly is a very nice library that lets us create eye-appealing maps. Therefore we will use plotly's express to create the maps. This time, we will not only look at cases and deaths, but also the number of recovered cases.

# Plotly
import matplotlib.ticker as ticker
import plotly.express as pex
from plotly.offline import plot

# New dataset will cases along with countries latitude and longitudes
confirmed_cases = pd.read_csv('https://raw.githubusercontent.com/CSSEGISandData/COVID-19/master/csse_covid_19_data/csse_covid_19_time_series/time_series_covid19_confirmed_global.csv')
confirmed_deaths = pd.read_csv('https://raw.githubusercontent.com/CSSEGISandData/COVID-19/master/csse_covid_19_data/csse_covid_19_time_series/time_series_covid19_deaths_global.csv')
confirmed_recovered = pd.read_csv('https://raw.githubusercontent.com/CSSEGISandData/COVID-19/master/csse_covid_19_data/csse_covid_19_time_series/time_series_covid19_recovered_global.csv')

confirmed_cases.head()

	Province/State	Country/Region	Lat	Long	1/22/20	1/23/20	1/24/20	1/25/20	1/26/20	1/27/20	...	12/18/20	12/19/20	12/20/20	12/21/20	12/22/20	12/23/20	12/24/20	12/25/20	12/26/20	12/27/20
0	NaN	Afghanistan	33.93911	67.709953	0	0	0	0	0	0	...	49621	49681	49817	50013	50190	50433	50655	50810	50886	51039
1	NaN	Albania	41.15330	20.168300	0	0	0	0	0	0	...	52004	52542	53003	53425	53814	54317	54827	55380	55755	56254
2	NaN	Algeria	28.03390	1.659600	0	0	0	0	0	0	...	94371	94781	95203	95659	96069	96549	97007	97441	97857	98249
3	NaN	Andorra	42.50630	1.521800	0	0	0	0	0	0	...	7519	7560	7577	7602	7633	7669	7699	7756	7806	7821
4	NaN	Angola	-11.20270	17.873900	0	0	0	0	0	0	...	16562	16626	16644	16686	16802	16931	17029	17099	17149	17240

5 rows × 345 columns

5.1 TreeMap of the world

Now we have access to three datasets for cases, deaths, and recovered, we can begin doing our visualization. As you can see, this dataset is an extension of the previous dataset because they include latitude and longitude. We can use this information to create maps.

We will begin by organizing the data frames. The datasets contain Mainland China as a country which should just be China. So, we will rename the columns, replace all entries for 'Mainland china' with just 'China', get rid of all the null values or empty entries, aggregate the results by summing the cases, deaths, and recovered cases. We will then create a simple data frame with these counts and plot them in a TreeMap.

# Rename some columns
confirmed_cases.rename(columns = {"Province/State": "State", 
                                 "Country/Region": "Country"}, inplace = True)
confirmed_deaths.rename(columns = {"Province/State": "State", 
                                 "Country/Region": "Country"}, inplace = True) 
confirmed_recovered.rename(columns = {"Province/State": "State", 
                                 "Country/Region": "Country"}, inplace = True) 

# Replace country that have Mainland China to just China
confirmed_cases['Country'].replace('Mainland China', 'China', inplace=True)
confirmed_deaths['Country'].replace('Mainland China', 'China', inplace=True)
confirmed_recovered['Country'].replace('Mainland China', 'China', inplace=True)

# Get rid of empty data
confirmed_cases[['State']] = confirmed_cases[['State']].fillna('')
confirmed_deaths[['State']] = confirmed_deaths[['State']].fillna('')
confirmed_recovered[['State']] = confirmed_recovered[['State']].fillna('')
confirmed_cases.fillna(0, inplace=True)
confirmed_deaths.fillna(0, inplace=True)
confirmed_recovered.fillna(0, inplace=True)

# Aggregate all the cases so we can see total sum for all countries
cases_count = confirmed_cases.iloc[:, 4:].sum().max()
deaths_count = confirmed_deaths.iloc[:, 4:].sum().max()
recovered_count = confirmed_recovered.iloc[:, 4:].sum().max()

# Print the info 
print('=============================')
print('Confirmed cases:', cases_count)
print('=============================')
print('Confirmed deaths:', deaths_count)
print('=============================')
print('Confirmed recovered:', recovered_count)
print('=============================')

# Make a dataframe consisting of the counts
counts_df = pd.DataFrame({
    'Total cases': [cases_count],
    'Total deaths': [deaths_count],
    'Total recovered': [recovered_count],
    'Active cases': [cases_count - deaths_count - recovered_count]
})

print("Dataframe with the counts-\n")
print(counts_df)

# We can convert all rows to columns in long format using melt
counts_long_df = counts_df.melt(value_vars=['Active cases', 'Total deaths', 'Total recovered'],
                               var_name='status', value_name='count')
counts_long_df['Upper'] = 'Total cases'
print('==============================================================')
print('Dataframe with counts as rows')
print(counts_long_df)

=============================
Confirmed cases: 80783674
=============================
Confirmed deaths: 1764863
=============================
Confirmed recovered: 47320022
=============================
Dataframe with the counts-

   Total cases  Total deaths  Total recovered  Active cases
0     80783674       1764863         47320022      31698789
==============================================================
Dataframe with counts as rows
            status     count        Upper
0     Active cases  31698789  Total cases
1     Total deaths   1764863  Total cases
2  Total recovered  47320022  Total cases

Now that we have our simple dataframe, we can plot the counts in a treemap.

# Make a treemap for the cases using plotly express
fig = pex.treemap(counts_long_df, path=['status'], values='count',
                 color_discrete_sequence=['red', 'blue', 'green'],
                 template='plotly_dark')
fig.show()

Analysis

In the map, there are way more recovered cases then the active cases which is great news. More people have recovered than the active cases. However, the blue box is amounts to active cases which is small fragment of total cases. If we plot a map with total cases, it will surely dominate the map. Total deaths is small amount on the map which makes sense. We can also hover over each box to get exact count of the box.

In the next section, we will create an single plot for the total cases, recovered cases, active cases, and total deaths.

5.2 Overall Plot of the World

from matplotlib.ticker import NullFormatter
from matplotlib.dates import MonthLocator, DateFormatter

# Sum all of the columns except the first four
confirmed_cases_count = confirmed_cases.iloc[:, 4:].sum(axis=0)
confirmed_deaths_count = confirmed_deaths.iloc[:, 4:].sum(axis=0)
confirmed_recovered_count = confirmed_recovered.iloc[:, 4:].sum(axis=0)
confirmed_active_count = confirmed_cases_count-confirmed_deaths_count-confirmed_recovered_count

# Using Seaborn to plot
fig, ax = plt.subplots(figsize=(20, 10))
sns.lineplot(x=confirmed_cases_count.index, y=confirmed_cases_count, sort=False, linewidth=3)
sns.lineplot(x=confirmed_deaths_count.index, y=confirmed_deaths_count, sort=False, linewidth=3)
sns.lineplot(x=confirmed_recovered_count.index, y=confirmed_recovered_count, sort=False, linewidth=3)
sns.lineplot(x=confirmed_active_count.index, y=confirmed_active_count, sort=False, linewidth=3)

# Set xlabel, y label, and titles
plt.title("Covid Confirmed/Deaths/Recovered/Active plot worldwide", fontsize=16, fontweight='bold')
plt.xticks(rotation=90)
plt.xlabel('Date', fontsize=13, fontweight='bold')
plt.ylabel('Number of cases', fontsize=13, fontweight='bold')
ax.legend(['Confirmed Cases', 'Confirmed Deaths', 'Confirmed Recovered', 'Active cases'])

# Set the date format to show data by months
ax.xaxis.set_major_locator(MonthLocator())
ax.xaxis.set_minor_locator(MonthLocator(bymonthday=2))
ax.xaxis.set_major_formatter(NullFormatter())
ax.xaxis.set_minor_formatter(DateFormatter('%b'))

plt.show()

The plot looks very similar to the other plots we have seen. In this plot, we can see one more line (purple) which represents the active cases. In the plot everything seems to be increasing. Confirmed cases seems to be increasing exponentially. Deaths seem to be below 1 million which makes sense, but it is still increasing through time.

5.3 Cases GeoMap of the world

We will do a final visualization which shows cases for most countries around the world. Since there are too many countries, I made a threshold of 50000 cases. Only countries with more than 50 thousand confirmed cases will be included in the map. This makes the map more smooth and easy to visualize.

We could use Folium to do a map. However, since I used plotly in early map, I am using it here as well for consistency. Plotly is also easier to use for maps. Regardless, both libraries portray the same relevant information.

# World plot using latitude and longitude
confirmed_cases_agg = confirmed_cases.groupby('Country').sum().reset_index()
confirmed_cases_agg.loc[:, ['Lat', 'Long']] = confirmed_cases.groupby('Country').mean().reset_index().loc[:, ['Lat', 'Long']]

# Plot only countries with more than 50,000 cases
MIN_CASES = 50000
confirmed_cases_agg['totals'] = confirmed_cases.iloc[:, 4:].sum(axis=1)
confirmed_cases_agg[['Country', 'Lat', 'Long', 'totals']].head()

confirmed_cases_agg = confirmed_cases_agg[confirmed_cases_agg.iloc[:, 3:].max(axis=1) > MIN_CASES]

# Change the cases to long format
confirmed_cases_agg_long = pd.melt(confirmed_cases_agg,
                                   id_vars=confirmed_cases_agg.iloc[:, :3],
                                   var_name='date',
                                   value_vars=confirmed_cases_agg.iloc[:, 3:],
                                   value_name='confirmed_date_cases')

# Use plotly to visualize the cases
fig = pex.scatter_geo(confirmed_cases_agg_long,
                     lat="Lat", lon="Long", color="Country",
                     hover_name="Country", size="confirmed_date_cases",
                     size_max=50, animation_frame="date",
                     template='plotly_dark', projection="natural earth",
                     title="Cases over time")

fig.show()

This map contains a visual map of the world. We can hit the play button to play the timeline of the world as cases increase throughout the world. The countries section is a scrollable which has all the countries in alphabetical order.

6. Conclusion

The novel coronavirus has had a massive impact on our world in the year 2020. Although we are at the end of 2020, the cases are still accumulating throughout the world. In this project, we looked at datasets for the United States and did some visualizations for it. We then proceeded with datasets for the entire world and explored the cases and deaths for the whole world. We found out that the United States has been the predominant country in the number of cases as well as the deaths. We did some predicting which showed an increase in cases in the next 10 days. This shows how terrible the United States has been this year because of Covid. In our maps, we did see that more people than the number of active cases have recovered from the disease which is good news. The project ended with an overall plot and a map showing all the cases through a map timeline.

The files for some of the dataset as well as the visual maps can be found in my github repository https://github.com/keshab101/keshab101.github.io

Further Exploration

The analysis is not entirely complete here. This is because the datasets are updated every day as new cases arise. There is room for more analysis using other attributes. There is also room for better predictions of future cases. Recently, we have heard of vaccine approval for coronavirus which might bring a tremendous change to the datasets. We can do further explorations of how vaccines are affecting the cases and how effective they are at reducing the cases. Moreover, we can see include other factors of a country such as their healthcare system, the poverty rate, and their population to see if the cases truly reflect those factors.