Clustering of European Countries by Economic Indicators
Mekhroj Doliev
2026-02-01
Introduction
In this project I will try to group European countries based on their economic characteristics. The idea is to find which countries are similar to each other. I use data from World Bank (https://data.worldbank.org/).
The main questions: - How many groups (clusters) exist? - Which countries belong together? - What makes each group different?
Loading Libraries and Data
# install if needed
if (!require("cluster")) install.packages("cluster")
if (!require("factoextra")) install.packages("factoextra")
if (!require("corrplot")) install.packages("corrplot")
library(cluster)
library(factoextra)
library(corrplot)# data from World Bank 2022
europe <- data.frame(
Country = c("Albania", "Austria", "Belgium", "Bulgaria", "Croatia",
"Czech Republic", "Denmark", "Estonia", "Finland", "France",
"Germany", "Greece", "Hungary", "Ireland", "Italy",
"Latvia", "Lithuania", "Netherlands", "Norway", "Poland",
"Portugal", "Romania", "Slovakia", "Slovenia", "Spain",
"Sweden", "Switzerland", "United Kingdom"),
GDP_per_capita = c(6810, 52085, 49582, 13772, 18570, 27220, 67803, 28247,
50732, 40886, 48636, 20867, 18390, 103685, 34776, 21947,
24032, 57025, 106149, 18688, 24521, 15892, 21088, 28439,
29674, 56424, 93259, 45295),
Life_expectancy = c(76.5, 81.3, 81.9, 74.8, 78.1, 78.3, 81.4, 78.6, 82.0,
82.5, 80.6, 80.1, 76.2, 82.0, 82.9, 75.3, 75.7, 81.4,
83.2, 77.0, 81.1, 74.2, 77.0, 80.6, 83.0, 83.0, 84.0, 80.4),
Unemployment = c(11.0, 4.8, 5.6, 4.3, 7.0, 2.2, 4.5, 5.6, 6.8, 7.3, 3.1,
12.4, 3.6, 4.5, 8.1, 6.9, 5.9, 3.5, 3.2, 2.9, 6.0, 5.6,
6.1, 4.0, 12.9, 7.5, 4.3, 3.7),
Education = c(58, 89, 79, 79, 67, 65, 82, 70, 93, 67, 72, 143, 52, 80, 64,
88, 72, 90, 84, 67, 66, 52, 60, 82, 93, 78, 63, 62),
Internet_users = c(79, 93, 92, 80, 82, 88, 98, 92, 93, 86, 93, 83, 89, 92,
85, 90, 87, 93, 98, 87, 84, 84, 90, 89, 94, 95, 96, 97),
CO2_emissions = c(1.8, 7.1, 8.0, 5.7, 4.3, 9.3, 5.1, 8.4, 7.5, 4.6, 8.1,
5.7, 4.9, 7.7, 5.3, 3.8, 4.5, 8.8, 7.5, 8.1, 4.4, 3.7,
5.8, 6.1, 5.1, 3.8, 4.0, 5.2)
)
rownames(europe) <- europe$Country# quick look at data
head(europe)## Country GDP_per_capita Life_expectancy Unemployment
## Albania Albania 6810 76.5 11.0
## Austria Austria 52085 81.3 4.8
## Belgium Belgium 49582 81.9 5.6
## Bulgaria Bulgaria 13772 74.8 4.3
## Croatia Croatia 18570 78.1 7.0
## Czech Republic Czech Republic 27220 78.3 2.2
## Education Internet_users CO2_emissions
## Albania 58 79 1.8
## Austria 89 93 7.1
## Belgium 79 92 8.0
## Bulgaria 79 80 5.7
## Croatia 67 82 4.3
## Czech Republic 65 88 9.3summary(europe[,-1])## GDP_per_capita Life_expectancy Unemployment Education
## Min. : 6810 Min. :74.20 Min. : 2.200 Min. : 52.00
## 1st Qu.: 21033 1st Qu.:77.00 1st Qu.: 3.925 1st Qu.: 64.75
## Median : 29056 Median :80.60 Median : 5.600 Median : 72.00
## Mean : 40160 Mean :79.75 Mean : 5.832 Mean : 75.61
## 3rd Qu.: 51070 3rd Qu.:82.00 3rd Qu.: 6.925 3rd Qu.: 82.50
## Max. :106149 Max. :84.00 Max. :12.900 Max. :143.00
## Internet_users CO2_emissions
## Min. :79.00 Min. :1.800
## 1st Qu.:85.75 1st Qu.:4.475
## Median :90.00 Median :5.500
## Mean :89.61 Mean :5.868
## 3rd Qu.:93.00 3rd Qu.:7.550
## Max. :98.00 Max. :9.300Correlation Check
Before clustering lets see if variables are related.
cor_matrix <- cor(europe[,-1])
corrplot(cor_matrix, method = "number", type = "upper")
GDP and life expectancy are positively correlated which makes sense - richer countries usually have better healthcare.
Data Standardization
Variables have different scales (GDP in thousands vs percentages) so we need to standardize.
data_scaled <- scale(europe[,-1])
head(data_scaled)## GDP_per_capita Life_expectancy Unemployment Education
## Albania -1.2598410 -1.1125061 1.91125058 -0.9912533
## Austria 0.4504572 0.5287762 -0.38172178 0.7539959
## Belgium 0.3559045 0.7339365 -0.08585438 0.1910123
## Bulgaria -0.9968461 -1.6937935 -0.56663891 0.1910123
## Croatia -0.8155979 -0.5654120 0.43191357 -0.4845681
## Czech Republic -0.4888374 -0.4970252 -1.34329083 -0.5971648
## Internet_users CO2_emissions
## Albania -1.9879612 -2.15864455
## Austria 0.6358799 0.65384756
## Belgium 0.4484626 1.13144056
## Bulgaria -1.8005440 -0.08907488
## Croatia -1.4257096 -0.83199733
## Czech Republic -0.3012062 1.82129711Finding Optimal Number of Clusters
Elbow Method
fviz_nbclust(data_scaled, kmeans, method = "wss") +
geom_vline(xintercept = 3, linetype = 2, color = "red")
Looks like elbow is around 3.
Silhouette Method
fviz_nbclust(data_scaled, kmeans, method = "silhouette")
Silhouette suggests 2, but 3 also looks ok.
I will go with 3 clusters because it gives more interesting grouping.
K-Means Clustering
set.seed(123)
km <- kmeans(data_scaled, centers = 3, nstart = 25)
# cluster sizes
table(km$cluster)##
## 1 2 3
## 12 2 14# add cluster to data
europe$cluster <- as.factor(km$cluster)
# visualize
fviz_cluster(km, data = data_scaled, geom = "point",
ellipse.type = "convex") +
ggtitle("K-Means Clustering (k=3)")
Which Countries in Each Cluster?
for(i in 1:3) {
cat("Cluster", i, ":\n")
cat(europe$Country[europe$cluster == i], sep = ", ")
cat("\n\n")
}## Cluster 1 :
## Albania, Bulgaria, Croatia, France, Hungary, Italy, Latvia, Lithuania, Poland, Portugal, Romania, Slovakia
##
## Cluster 2 :
## Greece, Spain
##
## Cluster 3 :
## Austria, Belgium, Czech Republic, Denmark, Estonia, Finland, Germany, Ireland, Netherlands, Norway, Slovenia, Sweden, Switzerland, United KingdomCluster Profiles
Lets see average values for each cluster:
cluster_means <- aggregate(europe[,2:7], by = list(Cluster = europe$cluster), FUN = mean)
cluster_means[,-1] <- round(cluster_means[,-1], 1)
print(cluster_means)## Cluster GDP_per_capita Life_expectancy Unemployment Education Internet_users
## 1 1 21614.3 77.6 6.2 66.0 85.2
## 2 2 25270.5 81.6 12.7 118.0 88.5
## 3 3 58184.4 81.3 4.5 77.8 93.5
## CO2_emissions
## 1 4.7
## 2 5.4
## 3 6.9Looking at results: - One cluster has highest GDP (rich countries
like Norway, Switzerland, Ireland) - One has medium values (most Western
European countries)
- One has lower GDP but still developing (Eastern European
countries)
Hierarchical Clustering
Lets compare with another method.
dist_matrix <- dist(data_scaled)
hc <- hclust(dist_matrix, method = "ward.D2")
plot(hc, main = "Dendrogram", cex = 0.7)
rect.hclust(hc, k = 3)
hc_clusters <- cutree(hc, k = 3)
europe$hc_cluster <- as.factor(hc_clusters)
# compare with kmeans
table(KMeans = europe$cluster, Hierarchical = europe$hc_cluster)## Hierarchical
## KMeans 1 2 3
## 1 11 1 0
## 2 0 0 2
## 3 0 14 0Results are similar but not exactly the same. Some countries are on the border between clusters.
PAM Clustering
PAM uses medoids instead of means - its more robust to outliers.
pam_result <- pam(data_scaled, k = 3)
europe$pam_cluster <- as.factor(pam_result$clustering)
# medoids (representative countries)
europe$Country[pam_result$medoids]## [1] "Austria" "Belgium" "Bulgaria" "Croatia"
## [5] "Czech Republic" "Denmark" "Estonia" "Finland"
## [9] "France" "Germany" "Greece" "Hungary"
## [13] "Ireland" "Italy" "Latvia" "Lithuania"
## [17] "Netherlands" "Norway" "Poland" "Portugal"
## [21] "Romania" "Slovakia" "Slovenia" "Spain"
## [25] "Sweden" "Switzerland" "United Kingdom"fviz_cluster(pam_result, data = data_scaled, geom = "point",
ellipse.type = "convex") +
ggtitle("PAM Clustering")
Validation - Silhouette Analysis
sil <- silhouette(km$cluster, dist_matrix)
fviz_silhouette(sil)## cluster size ave.sil.width
## 1 1 12 0.30
## 2 2 2 0.18
## 3 3 14 0.33
Average silhouette width is around 0.3 which is not great but acceptable. Some countries have low silhouette meaning they could belong to different cluster.
Conclusions
What I found:
European countries can be divided into roughly 3 groups based on economic indicators
The richest group includes Norway, Switzerland, Ireland, Denmark - these have GDP over $50k
Middle group has most Western European countries like France, Germany, UK
Third group is mostly Eastern European countries that are still developing economically
All three methods (K-means, Hierarchical, PAM) gave similar results which is good
Limitations: - Only used data from one year - 6 variables might not be enough - Some countries are between clusters (like Greece, Spain)
References
- World Bank: https://data.worldbank.org/
- Kassambara (2017) - Cluster Analysis in R
sessionInfo()## R version 4.5.1 (2025-06-13)
## Platform: aarch64-apple-darwin20
## Running under: macOS Tahoe 26.2
##
## Matrix products: default
## BLAS: /Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/lib/libRblas.0.dylib
## LAPACK: /Library/Frameworks/R.framework/Versions/4.5-arm64/Resources/lib/libRlapack.dylib; LAPACK version 3.12.1
##
## locale:
## [1] en_US.UTF-8/en_US.UTF-8/en_US.UTF-8/C/en_US.UTF-8/en_US.UTF-8
##
## time zone: Europe/Warsaw
## tzcode source: internal
##
## attached base packages:
## [1] stats graphics grDevices utils datasets methods base
##
## other attached packages:
## [1] corrplot_0.95 factoextra_1.0.7 ggplot2_4.0.0 cluster_2.1.8.1
##
## loaded via a namespace (and not attached):
## [1] gtable_0.3.6 jsonlite_2.0.0 dplyr_1.1.4 compiler_4.5.1
## [5] ggsignif_0.6.4 tidyselect_1.2.1 Rcpp_1.1.0 tidyr_1.3.1
## [9] jquerylib_0.1.4 scales_1.4.0 yaml_2.3.10 fastmap_1.2.0
## [13] R6_2.6.1 ggpubr_0.6.2 labeling_0.4.3 generics_0.1.4
## [17] Formula_1.2-5 knitr_1.50 backports_1.5.0 ggrepel_0.9.6
## [21] tibble_3.3.0 car_3.1-3 bslib_0.9.0 pillar_1.11.1
## [25] RColorBrewer_1.1-3 rlang_1.1.6 broom_1.0.10 cachem_1.1.0
## [29] xfun_0.53 sass_0.4.10 S7_0.2.0 cli_3.6.5
## [33] withr_3.0.2 magrittr_2.0.4 digest_0.6.37 grid_4.5.1
## [37] rstudioapi_0.17.1 lifecycle_1.0.4 vctrs_0.6.5 rstatix_0.7.3
## [41] evaluate_1.0.5 glue_1.8.0 farver_2.1.2 abind_1.4-8
## [45] carData_3.0-6 rmarkdown_2.30 purrr_1.1.0 tools_4.5.1
## [49] pkgconfig_2.0.3 htmltools_0.5.8.1