Clustering Countries (CIA Factbook)

# import useful libraries import numpy as np import pandas as pd import matplotlib.pyplot as plt import seaborn as sns

df = pd.read_csv('CIA_Country_Facts.csv')

# features of countries df.columns

# 227 countries with 20 attributes per country df.shape

# 5 countries with largest population # China and India have much larger populations than any other countries df.sort_values('Population',ascending=False)[['Country','Population']][0:5]

Countryobject

Populationint64

China

1313973713

India

1095351995

214

United States

298444215

Indonesia

245452739

Brazil

188078227

# histogram of population of countries # we set xlim so that we exclude China and India plt.figure(figsize=(10,8)) sns.histplot(data=df,x='Population',bins=100) plt.xlim(0,3e8);

# barplot of Region vs. GDP plt.figure(figsize=(12,8)) sns.barplot(data=df,x='Region',y='GDP ($ per capita)') plt.xticks(rotation=90);

# 5 countries with highest GDP per Capita df.sort_values('GDP ($ per capita)',ascending=False)[['Country','GDP ($ per capita)']][0:5]

Countryobject

GDP ($ per capita)float64

121

Luxembourg

55100

154

Norway

37800

214

United States

37800

Bermuda

36000

Cayman Islands

35000

# scatter plot of GDP per Capita vs. Phones, hue by Region # it seems that as GDP per Capita increases, more population tend to have phones plt.figure(figsize=(12,8)) sns.scatterplot(data=df,x='GDP ($ per capita)',y='Phones (per 1000)',hue='Region');

# barplot of Region vs. Phones # again, Western Europe and NA have highest numbers for Phones plt.figure(figsize=(12,8)) sns.barplot(data=df,x='Region',y='Phones (per 1000)') plt.xticks(rotation=90);

# country where everyone has a phone df.sort_values('Phones (per 1000)',ascending=False)[['Country','Phones (per 1000)','GDP ($ per capita)']][0:1]

Countryobject

Phones (per 1000)float64

138

Monaco

1035.6

# rich country, but not that many phones df.sort_values('GDP ($ per capita)',ascending=False)[['Country','Phones (per 1000)','GDP ($ per capita)']][0:1]

Countryobject

Phones (per 1000)float64

121

Luxembourg

515.4

# scatter plot of GDP per Capita vs. Literacy Rate, hue by Region plt.figure(figsize=(12,8)) sns.scatterplot(data=df,x='GDP ($ per capita)',y='Literacy (%)',hue='Region') plt.hlines(y=70,xmin=0,xmax=50000,colors='red');

# scatter plot of GDP per Capita vs. Infant Mortality rate, hue by Region plt.figure(figsize=(12,8)) sns.scatterplot(data=df,x='GDP ($ per capita)',y='Infant mortality (per 1000 births)',hue='Region');

# scatter plot of GDP per Capita vs. Literacy Rate, hue by Region plt.figure(figsize=(12,8)) sns.scatterplot(data=df,x='GDP ($ per capita)',y='Literacy (%)',hue='Region');

# scatter plot of GDP per Capita vs. Death Rate, hue by Region plt.figure(figsize=(12,8)) sns.scatterplot(data=df,x='GDP ($ per capita)',y='Deathrate',hue='Region');

sns.clustermap(df.corr())

# number of columns are missing values df.isnull().sum()

# We start with Climate column, as it is missing 22 values df[df['Climate'].isnull()]

Countryobject

Angola4.5%

Canada4.5%

20 others90.9%

Regionobject

WESTERN EUROPE 31.8%

SUB-SAHARAN AFRICA 13.6%

7 others54.5%

Angola

SUB-SAHARAN AFRICA

Canada

NORTHERN AMERICA

Croatia

EASTERN EUROPE

Faroe Islands

WESTERN EUROPE

Gibraltar

WESTERN EUROPE

101

Italy

WESTERN EUROPE

115

Lebanon

NEAR EAST

118

Libya

NORTHERN AFRICA

120

Lithuania

BALTICS

121

Luxembourg

WESTERN EUROPE

# since climate has close relation with country's region, we use region's mean climate to fill NA mean_climate = df.groupby('Region')['Climate'].transform('mean') df['Climate'] = df['Climate'].fillna(mean_climate)

# although not strong, there seems correlation between literacy rate and region of a country # hence we do the same process as above mean_literacy = df.groupby('Region')['Literacy (%)'].transform('mean') df['Literacy (%)'] = df['Literacy (%)'].fillna(mean_literacy)

# countries with missing agriculture values are mostly small Islands # these countries simply do not have enough lands for agriculutre, so we fill NA with 0 df[df['Agriculture'].isnull()]['Country']

df['Agriculture'] = df['Agriculture'].fillna(0)

# fill empty entries with 0 df['Industry'] = df['Industry'].fillna(0) df['Service'] = df['Service'].fillna(0)

# we only have few countries now df.isnull().sum()

# since only few countries are still missing data, we will simply drop these countries df = df.dropna(axis=0)

df.isnull().sum()

# we have dropped 9 countries df.shape

# Country name is more of an index rather than a feature X = df.drop('Country',axis=1)

# Create dummy features for Region column X = pd.get_dummies(X) X

Populationint64

7026 - 1313973713

Area (sq. mi.)int64

2 - 17075200

31056997

647500

3581655

28748

32930091

2381740

57794

199

71201

468

12127071

1246700

13477

102

69108

443

39921833

2766890

2976372

29800

from sklearn.preprocessing import StandardScaler

scaler = StandardScaler() scaled_X = scaler.fit_transform(X) scaled_X

from sklearn.cluster import KMeans

ssd = [] for k in range(2,40): kmeans = KMeans(n_clusters=k) kmeans.fit(scaled_X) # inertia_ is sum of squared distances of samples to their closest cluster center ssd.append(kmeans.inertia_)

plt.plot(range(2,40),ssd,'o-');

pd.Series(ssd).diff().plot(kind='bar');

final_model = KMeans(n_clusters=5) final_model.fit(scaled_X)

final_model.labels_

# ISO code of every country iso = pd.read_csv('country_iso_codes.csv')

iso.head()

Countryobject

ISO Codeobject

Afghanistan

AFG

Akrotiri and Dhekelia – See United Kingdom, The

Åland Islands

ALA

Albania

ALB

Algeria

DZA

iso.set_index('Country')

ISO Codeobject

MMR1%

AUS0.7%

278 others98.3%

Afghanistan

AFG

Akrotiri and Dhekelia – See United Kingdom, The

Åland Islands

ALA

Albania

ALB

Algeria

DZA

American Samoa

ASM

Andorra

AND

Angola

AGO

Anguilla

AIA

Antarctica [a]

ATA

iso.set_index('Country')['ISO Code'].to_dict()

# create a dictionary for mapping iso_map = iso.set_index('Country')['ISO Code'].to_dict() df['ISO'] = df['Country'].map(iso_map)

df['Cluster'] = final_model.labels_

import plotly.express as px

df.columns

fig = px.choropleth(df, locations="ISO", color="Cluster", # lifeExp is a column of gapminder hover_name="Country", # column to add to hover information ) fig.show()