Pokemon
Advaith
2025-12-04
Issue Description
The Pokémon universe assigns each creature a set of combat stats (HP,
Attack, Defense, Special Attack, Special Defense, Speed) along with
physical characteristics such as height, weight, and base
experience.
In this project, I use the TidyTuesday Pokémon dataset to:
Understand how combat stats relate to each other and to physical
attributes
Build a simple statistical model to predict HP from other
variables
Identify Pokémon whose observed HP is much higher than expected
(potentially “stronger than they look”)
This analysis shows how multivariate modeling can reveal non-obvious
patterns in game design.
Questions
How are Pokémon combat stats (HP, Attack, Defense, etc.)
distributed, and how strongly are they correlated?
Can we predict a Pokémon’s HP using other statistics and physical
attributes (height, weight, base experience, speed, attack,
defense)?
Which Pokémon have much higher HP than expected, given their
other traits, and could be considered “over-tanky”?
Predict Pokémon HP using multivariate regression
Cluster Pokémon into combat archetypes
(offensive/defensive/speedy)
Data Source
Data: TidyTuesday – 2025-04-01 Pokémon week, based on a curated
dataset of Pokémon attributes. https://raw.githubusercontent.com/rfordatascience/tidytuesday/main/data/2025/2025-04-01/pokemon_df.csv”
#Load Packages
# Packages
library(tidytuesdayR)
## Warning: package 'tidytuesdayR' was built under R version 4.5.2
## Warning: package 'readr' was built under R version 4.5.2
## Warning: package 'dplyr' was built under R version 4.5.2
##
## Attaching package: 'dplyr'
## The following objects are masked from 'package:stats':
##
## filter, lag
## The following objects are masked from 'package:base':
##
## intersect, setdiff, setequal, union
## Warning: package 'ggplot2' was built under R version 4.5.2
## Warning: package 'GGally' was built under R version 4.5.2
library(broom)
library(stringr)
## Warning: package 'stringr' was built under R version 4.5.2
## Warning: package 'tidyverse' was built under R version 4.5.1
## Warning: package 'tibble' was built under R version 4.5.1
## Warning: package 'tidyr' was built under R version 4.5.1
## Warning: package 'purrr' was built under R version 4.5.1
## Warning: package 'forcats' was built under R version 4.5.1
## Warning: package 'lubridate' was built under R version 4.5.2
## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ forcats 1.0.0 ✔ tibble 3.3.0
## ✔ lubridate 1.9.4 ✔ tidyr 1.3.1
## ✔ purrr 1.1.0
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()
## ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors
tuesdata <- tidytuesdayR::tt_load("2025-04-01")
## ---- Compiling #TidyTuesday Information for 2025-04-01 ----
## --- There is 1 file available ---
##
##
## ── Downloading files ───────────────────────────────────────────────────────────
##
## 1 of 1: "pokemon_df.csv"
pokemon_df <- tuesdata$pokemon_df
Documentation
The dataset is described in the TidyTuesday README and in several
blog posts analyzing this week’s data.
Key columns (from the screenshot you gave plus documentation): id –
unique Pokémon ID pokemon – name species_id – species ID height, weight
– physical dimensions base_experience – experience yield type_1, type_2
– primary and secondary type hp, attack, defense, special_attack,
special_defense, speed – combat stats color_1, color_2, color_f – color
codes egg_group_1, egg_group_2 – breeding groups generation_id –
generation number
Description of the Data
## spc_tbl_ [949 × 22] (S3: spec_tbl_df/tbl_df/tbl/data.frame)
## $ id : num [1:949] 1 2 3 4 5 6 7 8 9 10 ...
## $ pokemon : chr [1:949] "bulbasaur" "ivysaur" "venusaur" "charmander" ...
## $ species_id : num [1:949] 1 2 3 4 5 6 7 8 9 10 ...
## $ height : num [1:949] 0.7 1 2 0.6 1.1 1.7 0.5 1 1.6 0.3 ...
## $ weight : num [1:949] 6.9 13 100 8.5 19 90.5 9 22.5 85.5 2.9 ...
## $ base_experience: num [1:949] 64 142 236 62 142 240 63 142 239 39 ...
## $ type_1 : chr [1:949] "grass" "grass" "grass" "fire" ...
## $ type_2 : chr [1:949] "poison" "poison" "poison" NA ...
## $ hp : num [1:949] 45 60 80 39 58 78 44 59 79 45 ...
## $ attack : num [1:949] 49 62 82 52 64 84 48 63 83 30 ...
## $ defense : num [1:949] 49 63 83 43 58 78 65 80 100 35 ...
## $ special_attack : num [1:949] 65 80 100 60 80 109 50 65 85 20 ...
## $ special_defense: num [1:949] 65 80 100 50 65 85 64 80 105 20 ...
## $ speed : num [1:949] 45 60 80 65 80 100 43 58 78 45 ...
## $ color_1 : chr [1:949] "#78C850" "#78C850" "#78C850" "#F08030" ...
## $ color_2 : chr [1:949] "#A040A0" "#A040A0" "#A040A0" NA ...
## $ color_f : chr [1:949] "#81A763" "#81A763" "#81A763" NA ...
## $ egg_group_1 : chr [1:949] "monster" "monster" "monster" "monster" ...
## $ egg_group_2 : chr [1:949] "plant" "plant" "plant" "dragon" ...
## $ url_icon : chr [1:949] "//archives.bulbagarden.net/media/upload/7/7b/001MS6.png" "//archives.bulbagarden.net/media/upload/a/a0/002MS6.png" "//archives.bulbagarden.net/media/upload/0/07/003MS6.png" "//archives.bulbagarden.net/media/upload/7/7d/004MS6.png" ...
## $ generation_id : num [1:949] 1 1 1 1 1 1 1 1 1 1 ...
## $ url_image : chr [1:949] "https://raw.githubusercontent.com/HybridShivam/Pokemon/master/assets/images/001.png" "https://raw.githubusercontent.com/HybridShivam/Pokemon/master/assets/images/002.png" "https://raw.githubusercontent.com/HybridShivam/Pokemon/master/assets/images/003.png" "https://raw.githubusercontent.com/HybridShivam/Pokemon/master/assets/images/004.png" ...
## - attr(*, "spec")=
## .. cols(
## .. id = col_double(),
## .. pokemon = col_character(),
## .. species_id = col_double(),
## .. height = col_double(),
## .. weight = col_double(),
## .. base_experience = col_double(),
## .. type_1 = col_character(),
## .. type_2 = col_character(),
## .. hp = col_double(),
## .. attack = col_double(),
## .. defense = col_double(),
## .. special_attack = col_double(),
## .. special_defense = col_double(),
## .. speed = col_double(),
## .. color_1 = col_character(),
## .. color_2 = col_character(),
## .. color_f = col_character(),
## .. egg_group_1 = col_character(),
## .. egg_group_2 = col_character(),
## .. url_icon = col_character(),
## .. generation_id = col_double(),
## .. url_image = col_character()
## .. )
## - attr(*, "problems")=<externalptr>
summary(select(
pokemon_df,
height, weight, base_experience,
hp, attack, defense, special_attack, special_defense, speed
))
## height weight base_experience hp
## Min. : 0.100 Min. : 0.10 Min. : 36.0 Min. : 1.00
## 1st Qu.: 0.500 1st Qu.: 8.50 1st Qu.: 68.0 1st Qu.: 50.00
## Median : 1.000 Median : 28.80 Median :157.0 Median : 65.00
## Mean : 1.228 Mean : 66.21 Mean :150.5 Mean : 68.95
## 3rd Qu.: 1.500 3rd Qu.: 66.60 3rd Qu.:184.0 3rd Qu.: 80.00
## Max. :14.500 Max. :999.90 Max. :608.0 Max. :255.00
## attack defense special_attack special_defense
## Min. : 5.00 Min. : 5.00 Min. : 10.00 Min. : 20.00
## 1st Qu.: 55.00 1st Qu.: 50.00 1st Qu.: 50.00 1st Qu.: 50.00
## Median : 75.00 Median : 70.00 Median : 65.00 Median : 70.00
## Mean : 79.47 Mean : 74.07 Mean : 72.81 Mean : 72.22
## 3rd Qu.:100.00 3rd Qu.: 90.00 3rd Qu.: 95.00 3rd Qu.: 90.00
## Max. :190.00 Max. :230.00 Max. :194.00 Max. :230.00
## speed
## Min. : 5.00
## 1st Qu.: 45.00
## Median : 65.00
## Mean : 69.02
## 3rd Qu.: 90.00
## Max. :180.00
To get an overview of how the numeric combat stats relate, I create a
smaller numeric subset:
stats_cols <- c("hp", "attack", "defense",
"special_attack", "special_defense", "speed")
pokemon_stats <- pokemon_df |>
select(all_of(stats_cols))
GGally::ggpairs(pokemon_stats)
Cleaning and Preparation
The Pokémon data are already quite clean, but I still:
- Standardize the Pokémon name (optional, for nicer labels).
- Remove rows with missing values in key numeric columns.
- Create a modeling dataset focused on the variables relevant to HP
prediction.
pokemon_clean <- pokemon_df |>
mutate(
pokemon = str_to_title(pokemon)
) |>
filter(
!is.na(hp),
!is.na(attack),
!is.na(defense),
!is.na(speed),
!is.na(height),
!is.na(weight),
!is.na(base_experience)
)
summary(select(
pokemon_clean,
height, weight, base_experience,
hp, attack, defense, special_attack, special_defense, speed
))
## height weight base_experience hp
## Min. : 0.100 Min. : 0.10 Min. : 36.0 Min. : 1.00
## 1st Qu.: 0.500 1st Qu.: 8.50 1st Qu.: 68.0 1st Qu.: 50.00
## Median : 1.000 Median : 28.80 Median :157.0 Median : 65.00
## Mean : 1.228 Mean : 66.21 Mean :150.5 Mean : 68.95
## 3rd Qu.: 1.500 3rd Qu.: 66.60 3rd Qu.:184.0 3rd Qu.: 80.00
## Max. :14.500 Max. :999.90 Max. :608.0 Max. :255.00
## attack defense special_attack special_defense
## Min. : 5.00 Min. : 5.00 Min. : 10.00 Min. : 20.00
## 1st Qu.: 55.00 1st Qu.: 50.00 1st Qu.: 50.00 1st Qu.: 50.00
## Median : 75.00 Median : 70.00 Median : 65.00 Median : 70.00
## Mean : 79.47 Mean : 74.07 Mean : 72.81 Mean : 72.22
## 3rd Qu.:100.00 3rd Qu.: 90.00 3rd Qu.: 95.00 3rd Qu.: 90.00
## Max. :190.00 Max. :230.00 Max. :194.00 Max. :230.00
## speed
## Min. : 5.00
## 1st Qu.: 45.00
## Median : 65.00
## Mean : 69.02
## 3rd Qu.: 90.00
## Max. :180.00
Final Results
Q1 – Distribution and correlation of combat stats
pokemon_long <- pokemon_clean |>
select(all_of(stats_cols)) |>
pivot_longer(everything(), names_to = "stat", values_to = "value")
ggplot(pokemon_long,
aes(x = value)) +
geom_histogram(bins = 30, fill = "steelblue", color = "white") +
facet_wrap(~ stat, scales = "free") +
labs(
title = "Distributions of Pokémon Combat Stats",
x = "Stat value",
y = "Count"
)
cor(select(pokemon_clean, all_of(stats_cols)))
## hp attack defense special_attack special_defense
## hp 1.0000000 0.4266479 0.26085534 0.3737768 0.3700109
## attack 0.4266479 1.0000000 0.43962628 0.3848831 0.2518813
## defense 0.2608553 0.4396263 1.00000000 0.2237726 0.5314911
## special_attack 0.3737768 0.3848831 0.22377262 1.0000000 0.4924357
## special_defense 0.3700109 0.2518813 0.53149106 0.4924357 1.0000000
## speed 0.1433248 0.3586703 -0.02229393 0.4471509 0.2116434
## speed
## hp 0.14332480
## attack 0.35867033
## defense -0.02229393
## special_attack 0.44715089
## special_defense 0.21164343
## speed 1.00000000
Q2-Predicting HP from other attributes
hp_model <- lm(
hp ~ height + weight + base_experience +
attack + defense + speed,
data = pokemon_clean
)
summary(hp_model)
##
## Call:
## lm(formula = hp ~ height + weight + base_experience + attack +
## defense + speed, data = pokemon_clean)
##
## Residuals:
## Min 1Q Median 3Q Max
## -62.225 -10.628 -2.452 8.080 114.688
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 55.283570 2.268826 24.367 < 2e-16 ***
## height 2.360488 0.650023 3.631 0.000297 ***
## weight 0.037077 0.006928 5.352 1.1e-07 ***
## base_experience 0.250225 0.012220 20.477 < 2e-16 ***
## attack 0.062068 0.024873 2.495 0.012753 *
## defense -0.234316 0.025455 -9.205 < 2e-16 ***
## speed -0.245044 0.025644 -9.556 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 18.24 on 942 degrees of freedom
## Multiple R-squared: 0.5087, Adjusted R-squared: 0.5056
## F-statistic: 162.6 on 6 and 942 DF, p-value: < 2.2e-16
## # A tibble: 1 × 12
## r.squared adj.r.squared sigma statistic p.value df logLik AIC BIC
## <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 0.509 0.506 18.2 163. 1.22e-141 6 -4098. 8213. 8252.
## # ℹ 3 more variables: deviance <dbl>, df.residual <int>, nobs <int>
par(mfrow = c(2, 2))
plot(hp_model)
Q3 – Identifying “over-tanky” Pokémon (large positive HP
residuals)
hp_results <- pokemon_clean |>
mutate(
hp_pred = fitted(hp_model),
hp_residual = hp - hp_pred
)
top_over_hp <- hp_results |>
arrange(desc(hp_residual)) |>
select(pokemon, type_1, type_2,
hp, hp_pred, hp_residual) |>
slice_head(n = 15)
top_over_hp
## # A tibble: 15 × 6
## pokemon type_1 type_2 hp hp_pred hp_residual
## <chr> <chr> <chr> <dbl> <dbl> <dbl>
## 1 Wobbuffet psychic <NA> 190 75.3 115.
## 2 Guzzlord dark dragon 223 113. 110.
## 3 Chansey normal <NA> 250 145. 105.
## 4 Alomomola water <NA> 165 70.6 94.4
## 5 Zygarde-Complete dragon ground 216 125. 90.6
## 6 Drifblim ghost flying 150 77.3 72.7
## 7 Solgaleo psychic steel 137 65.5 71.5
## 8 Lunala psychic ghost 137 65.6 71.4
## 9 Wailmer water <NA> 130 66.3 63.7
## 10 Munchlax normal <NA> 135 74.8 60.2
## 11 Slaking normal <NA> 150 89.9 60.1
## 12 Blissey normal <NA> 255 197. 57.5
## 13 Aurorus rock ice 123 69.7 53.3
## 14 Nihilego rock poison 109 55.7 53.3
## 15 Lanturn water electric 125 72.8 52.2
ggplot(top_over_hp,
aes(x = reorder(pokemon, hp_residual),
y = hp_residual,
fill = type_1)) +
geom_col() +
coord_flip() +
labs(
title = "Pokémon with Much Higher HP Than Expected",
x = "Pokémon",
y = "HP residual (observed - predicted)"
) +
theme(legend.position = "bottom")
Q4-Predict Pokémon HP using multivariate regression
hp_model <- lm(
hp ~ height + weight + base_experience +
attack + defense + special_attack +
special_defense + speed,
data = pokemon_clean
)
hp_model_summary <- summary(hp_model)
# Add predictions + residuals
pokemon_regression <- pokemon_clean |>
mutate(
hp_pred = predict(hp_model),
hp_residual = hp - hp_pred
)
# Top Pokémon with unusually high HP (positive residuals)
top_hp_outliers <- pokemon_regression |>
arrange(desc(hp_residual)) |>
select(pokemon, hp, hp_pred, hp_residual) |>
slice_head(n = 10)
##
## Call:
## lm(formula = hp ~ height + weight + base_experience + attack +
## defense + special_attack + special_defense + speed, data = pokemon_clean)
##
## Residuals:
## Min 1Q Median 3Q Max
## -64.480 -10.270 -2.138 7.829 111.923
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 59.071206 2.614573 22.593 < 2e-16 ***
## height 2.727561 0.655633 4.160 3.47e-05 ***
## weight 0.036278 0.006897 5.260 1.78e-07 ***
## base_experience 0.281054 0.015759 17.834 < 2e-16 ***
## attack 0.047388 0.026246 1.806 0.07131 .
## defense -0.235222 0.027157 -8.661 < 2e-16 ***
## special_attack -0.077753 0.025661 -3.030 0.00251 **
## special_defense -0.036156 0.032249 -1.121 0.26251
## speed -0.235174 0.025770 -9.126 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 18.14 on 940 degrees of freedom
## Multiple R-squared: 0.5147, Adjusted R-squared: 0.5106
## F-statistic: 124.6 on 8 and 940 DF, p-value: < 2.2e-16
## # A tibble: 10 × 4
## pokemon hp hp_pred hp_residual
## <chr> <dbl> <dbl> <dbl>
## 1 Guzzlord 223 111. 112.
## 2 Wobbuffet 190 79.1 111.
## 3 Chansey 250 155. 94.9
## 4 Alomomola 165 74.6 90.4
## 5 Zygarde-Complete 216 129. 87.1
## 6 Lunala 137 59.6 77.4
## 7 Solgaleo 137 61.4 75.6
## 8 Drifblim 150 77.5 72.5
## 9 Wailmer 130 66.0 64.0
## 10 Munchlax 135 73.7 61.3
Q5-Cluster Pokémon into combat archetypes
(offensive/defensive/speedy)
# Scale numeric combat stats
pokemon_scaled <- scale(pokemon_stats)
set.seed(123)
k3 <- kmeans(pokemon_scaled, centers = 3, nstart = 25)
pokemon_clustered <- pokemon_clean |>
mutate(cluster = factor(k3$cluster))
# Visualize clusters (Attack vs Defense)
cluster_plot <- ggplot(pokemon_clustered,
aes(x = attack, y = defense, color = cluster)) +
geom_point(alpha = 0.7, size = 2) +
labs(
title = "Pokémon Combat Archetypes ",
x = "Attack",
y = "Defense",
color = "Cluster"
) +
theme_minimal()
# View cluster sizes
cluster_sizes <- table(pokemon_clustered$cluster)
##
## 1 2 3
## 286 382 281
