カテゴリごとのモデル

Rで学ぶ中級回帰分析

Richie Cotton

Data Evangelist at DataCamp

4つのカテゴリ

unique(fish$species)

"Bream" "Roach" "Perch" "Pike"

データセットの分割

スマートな方法

base-R: split() + lapply()
dplyr: nest_by() + mutate()

シンプルな方法

bream <- fish %>% 
  filter(species == "Bream")
perch <- fish %>% 
  filter(species == "Perch")
pike <- fish %>% 
  filter(species == "Pike")
roach <- fish %>% 
  filter(species == "Roach")

4つのモデル

mdl_bream <- lm(mass_g ~ length_cm, data = bream)

Call:
lm(formula = mass_g ~ length_cm, data = bream)

Coefficients:
(Intercept)    length_cm  
   -1035.35        54.55

mdl_pike <- lm(mass_g ~ length_cm, data = pike)

Call:
lm(formula = mass_g ~ length_cm, data = pike)

Coefficients:
(Intercept)    length_cm  
   -1540.82        53.19

mdl_perch <- lm(mass_g ~ length_cm, data = perch)

Call:
lm(formula = mass_g ~ length_cm, data = perch)

Coefficients:
(Intercept)    length_cm  
    -619.18        38.91

mdl_roach <- lm(mass_g ~ length_cm, data = roach)

Call:
lm(formula = mass_g ~ length_cm, data = roach)

Coefficients:
(Intercept)    length_cm  
    -329.38        23.32

説明変数データ

explanatory_data <- tibble(
  length_cm = seq(5, 60, 5)
)

予測の作成

prediction_data_bream <- explanatory_data %>%
  mutate(
    mass_g = predict(mdl_bream, explanatory_data),
    species = "Bream"
  )

prediction_data_pike <- explanatory_data %>%
  mutate(
    mass_g = predict(mdl_perch, explanatory_data),
    species = "Perch"
  )

prediction_data_perch <- explanatory_data %>%
  mutate(
    mass_g = predict(mdl_pike, explanatory_data),
    species = "Pike"
  )

prediction_data_roach <- explanatory_data %>%
  mutate(
    mass_g = predict(mdl_roach, explanatory_data),
    species = "Roach"
  )

予測の可視化

ggplot(fish, aes(length_cm, mass_g, color = species)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE)

種別の長さと質量の散布図（交互作用あり）の画像

予測の追加

ggplot(fish,aes(length_cm, mass_g, color = species)) +
  geom_point() +
  geom_smooth(method = "lm", se = FALSE) +
  geom_point(data = prediction_data_bream, size = 3, shape = 15) +
  geom_point(data = prediction_data_perch, size = 3, shape = 15) +
  geom_point(data = prediction_data_pike, size = 3, shape = 15) +
  geom_point(data = prediction_data_roach, size = 3, shape = 15)

予測を重ねた種別の長さと質量の散布図の画像

決定係数

mdl_fish <- lm(mass_g ~ length_cm + species, data = fish)

mdl_fish %>% 
  glance() %>% 
  pull(adj.r.squared)

0.917

mdl_bream %>% glance() %>% pull(adj.r.squared)

0.874

mdl_perch %>% glance() %>% pull(adj.r.squared)

0.917

mdl_pike %>% glance() %>% pull(adj.r.squared)

0.941

mdl_roach %>% glance() %>% pull(adj.r.squared)

0.815

残差標準誤差

mdl_fish %>% 
  glance() %>% 
  pull(sigma)

mdl_bream %>% glance() %>% pull(sigma)

74.2

mdl_perch %>% glance() %>% pull(sigma)

mdl_pike %>% glance() %>% pull(sigma)

mdl_roach %>% glance() %>% pull(sigma)

38.2

練習しましょう！

Rで学ぶ中級回帰分析