data analysis for nyc flights data set

flight_example.R

library(tidymodels)
library(nycflights13)

set.seed(25213)
flight_data <- 
  flights %>% 
  mutate(
    delay = ifelse(arr_delay >= 30, "late", "on_time"),
    delay = factor(delay),
    date = as.Date(time_hour)
  ) %>% 
  select(dep_time, flight, origin, dest, air_time, distance, 
         date, delay, dest, time_hour) %>% 
  inner_join(weather %>% select(-month, -day, -hour, -year), 
             by = c("origin", "time_hour")) %>% 
  select(-time_hour) %>% 
  na.omit() %>% 
  sample_n(10000) %>%
  mutate_if(is.character, as.factor)

set.seed(48256)
data_split <- initial_split(flight_data, prop = 3/4)
train_data <- training(data_split)
holdout_data <- testing(data_split)

lr_model <- logistic_reg() %>% set_engine("glm")

flights_rec <- 
  recipe(delay ~ ., data = train_data) %>% 
  step_date(date, features = c("dow", "month")) %>% 
  step_holiday(date, holidays = timeDate::listHolidays("US")) %>% 
  update_role(date, new_role = "date") %>% 
  step_dummy(all_nominal(), -all_outcomes()) %>% 
  step_zv(all_predictors())

flights_wflow <- 
  workflow() %>% 
  add_model(lr_model) %>% 
  add_recipe(flights_rec)

flights_fit <- fit(flights_wflow, data = train_data)

flights_pred <- 
  predict(flights_fit, holdout_data, type = "prob") %>% 
  bind_cols(holdout_data %>% select(delay))

flights_pred %>% 
  roc_curve(truth = delay, .pred_late) %>% 
  autoplot()

roc_auc(flights_pred, truth = delay, .pred_late)