Skip to content

Will my flight be delayed? Built with R, this app predicts the delay of a flight between any two US cities.

License

Notifications You must be signed in to change notification settings

henryburkhardt/delayed-flights

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

delayed-flights

Will my flight be delayed? Built with R (the statistical programing language), this app predicts the delay of a flight between any two US cities.

A note about the data: the data used in this project is publicly available from the Bureau of Transportation Statistics website. I used only data from the first month of January 2017 because I didn't have the capacity to work with larger CSV files. However, this code is totally scalable to work with data from a full year, or even multiple years to give more accurate predictions.

Introduction

This markdown file has two purposes: (1) creating a useful dataframe from the raw data imported below and functions to query that data and (2) creating map visualizations from this data. The second file in this directory, “app.R” is a user interface that uses the functions from this markdown file and returns their results to a Shiny App.

A note on the UI: The user interface built in ./app.R relies on variables and functions that are defined in this markdown document. All the chunks in this markdown must be run before app.R will run properly.

The user interface (app.R) looks like this:

If the code below is not of interest, be sure to at least check out the super cool visualizations generated from it!

#installing packages
library(readr)
library(tidyverse)
library(ggplot2)
library(ggmap)
library(maps)
library(sf)
library(rnaturalearth)
library(rnaturalearthdata)

#set up API key to acces the Google geocoding service
register_google(key ="[insert your own google cloud API key here :) ]") 

#reading flight data 
fd <- read_csv("./data_sets/2017_Flight_data.csv")

#getting previoulsy compiled geodata
if (file.exists("./data_sets/geo_lookup_city.csv")){
  geo_lookup_city <- read_csv("./data_sets/geo_lookup_city.csv")
}

Creating a More Useful Dataframe

The goal of this new dataframe is to compute summary statistics for every single flight in the 2017_flight_data.csv file. The original file has over 400,000 rows (one for every flight in the one month period of the dataframe), the new dataframe will have one row for every flight between two cities. For example, every flight between Minneapolis, MN and Denver, CO in the original data will be condensed into one row, and the selected statistics will be summarized into the single row. Thus a probability and average delay time can be assigned to every flight. This new dataframe is called fd_heatmap_group (it’s purpose was originally to create a map of each unique flight path).

Creating New Dataframe and Grouping by uniqueID

#selecting the variable I want in the new DF
fd_heatmap <- select(fd,DepDelayMinutes,OriginCityName, DestCityName, OriginAirportID, DestAirportID, Cancelled, DepDel15, DepDelay, Cancelled, UniqueCarrier)

#creating a unique id number for every flight between two specific cities
fd_heatmap$uniqueID <- as.numeric(paste(fd_heatmap$OriginAirportID, fd_heatmap$DestAirportID, sep = ''))

#combining and summarizing the data using the tidyverse group_by() method
fd_heatmap_group <- fd_heatmap %>% group_by(uniqueID) %>% 
  summarise(
    avgDelay = mean(DepDelayMinutes, na.rm  = TRUE), #the average minutes of delay 
    OriginCityName = first(OriginCityName), #Origin City name 
    DestCityName = first(DestCityName),#Destination city name
    quantity = n(), # the number of flights in each group  
    noDelay = length(which(cur_data()$DepDelayMinutes==0)),#amount not delayed
    delay15 = sum(cur_data()$DepDel15, na.rm = TRUE)#amount delayed 15+ minutes
  )

#converting decimal to percent strings. this becomes important later when we want to return a probablity to the user in the form of a percent charecter string.
fd_heatmap_group$delayOdds <- paste(round((1-(fd_heatmap_group$noDelay/fd_heatmap_group$quantity))*100), "%", sep = "")
fd_heatmap_group$delayOddsRaw <- (fd_heatmap_group$noDelay/fd_heatmap_group$quantity)
fd_heatmap_group$delay15Odds <- paste(round((fd_heatmap_group$delay15/fd_heatmap_group$quantity)*100, 0), "%", sep = "")
fd_heatmap_group$popularity <- fd_heatmap_group$quantity/max(fd_heatmap_group$quantity)

Assigning Lon/Lat Coordinates to Every City

Because we eventually want to be creating maps of flights, the each airport defined in the fd data needs to be assigned a latitude and longitude. This will allow us to progamatically draw as many plane flights as we want. Google’s Geocoding API was selected as the best method. The for loop below calls the API (linked with my Google Cloud account) on all of the unique cities in the fd data.

#this chunk should only be run as nesecary. A csv file is saved after the first run
allCities <- unique(c(fd_heatmap$OriginCityName, fd_heatmap$DestCityName))

lat <- c()
lon <- c()

for (i in 1:length(allCities)) {
  geocode <- geocode(allCities[i]) #calls google API
  lon[i] <- as.numeric(geocode[2])
  lat[i] <- as.numeric(geocode[1])
}

geo_lookup_city <- as.data.frame(cbind(allCities, lon, lat)) 

#Fixing city names for the two instacnes that the Google API could not locate. 
geo_lookup_city[geo_lookup_city$allCities == "Montrose/Delta, CO", "lon"] <- as.numeric(geocode("Montrose, CO")[2])
geo_lookup_city[geo_lookup_city$allCities == "Montrose/Delta, CO", "lat"] <- as.numeric(geocode("Montrose, CO")[1])

geo_lookup_city[geo_lookup_city$allCities == "Guam, TT", "lon"] <- as.numeric(geocode("Guam")[2])
geo_lookup_city[geo_lookup_city$allCities == "GUam, TT", "lat"] <- as.numeric(geocode("Guam")[1])

write_csv(geo_lookup_city, file="./geo_lookup_city.csv") 
#api can be slow so results are saved as a csv, doesn't have to be run everytime

Assigning Lat/Lon Coordinates to Every Flight

Each city in fd_heatmap_group is assigned its corresponding lat and lon from the previously generated geo_lookup_city .

fd_heatmap_group$orgin_lon <- geo_lookup_city$lon[match(fd_heatmap_group$OriginCityName, geo_lookup_city$allCities)]
fd_heatmap_group$orgin_lat <- geo_lookup_city$lat[match(fd_heatmap_group$OriginCityName, geo_lookup_city$allCities)]

fd_heatmap_group$destination_lon <- geo_lookup_city$lon[match(fd_heatmap_group$DestCityName, geo_lookup_city$allCities)]
fd_heatmap_group$destination_lat <- geo_lookup_city$lat[match(fd_heatmap_group$DestCityName, geo_lookup_city$allCities)]

Creating a Subset for Continental US Flights

Now that the geodata of each city is known, fd_heatmap_group can be queried geographically. The data is subsested geographically here so that we can plot map objects that only include the contenital us. The data is compared against a “bounding box” of lat/lon coordinates around the us.

#Cordinate bounding box for us lon = 25, 49 / lat = -126, -66
fd_heatmap_group_continetnal <- fd_heatmap_group[
  between(fd_heatmap_group$orgin_lon, 25,49)&
  between(fd_heatmap_group$orgin_lat, -126,-66)&
  between(fd_heatmap_group$destination_lon, 25,49)&
  between(fd_heatmap_group$destination_lat, -126,-66),]

Creating Functions

flightUi()

This first function is the back bone of making this data useful. It’s meant to be used in the command line and prompts the user to enter a departure and arrival city name. It checks to make sure both inputs exist in the data, and then returns the row in fd_heatmap_group that corresponds to the given flight path.

#type flightUi() into the command line after running this chunk to test it out.

flightUi <- function(){
  deptCity <- readline(prompt="DEPARTURE CITY: ")
  if (!(deptCity %in% fd_heatmap_group$OriginCityName)){
    cat("city not found")
    return()
    }

  arrvCity <- readline(prompt="ARRIVAL CITY: ")
  if (!(arrvCity %in% fd_heatmap_group$DestCityName)){
    cat("city not found")
    return()
  }
  
  tripData <- fd_heatmap_group[(fd_heatmap_group$DestCityName==arrvCity)&(fd_heatmap_group$OriginCityName==deptCity),]
  
  return(tripData)
}

flightUi2()

This function is almost identical to flightUi() except that it takes the departure and arrival city as inputs to the function instead of gathering input through the command line. This function becomes incredible useful in the Shiny app, when our goal is to operate on inputs provided by a user through the UI.

flightUi2 <- function(deptCity,arrvCity){
  if (!(deptCity %in% fd_heatmap_group$OriginCityName)){
    cat("city not found")
    return()
    }
  if (!(arrvCity %in% fd_heatmap_group$DestCityName)){
    cat("city not found")
    return()
  }
  
  tripData <- fd_heatmap_group[(fd_heatmap_group$DestCityName==arrvCity)&(fd_heatmap_group$OriginCityName==deptCity),]
  
  return(tripData)
}

flightUi2("Minneapolis, MN","Denver, CO")
## # A tibble: 1 x 15
##     uniqueID avgDelay OriginCityName  DestCityName quantity noDelay delay15
##        <dbl>    <dbl> <chr>           <chr>           <int>   <int>   <dbl>
## 1 1348711292     13.7 Minneapolis, MN Denver, CO        453     252     101
## # ... with 8 more variables: delayOdds <chr>, delayOddsRaw <dbl>,
## #   delay15Odds <chr>, popularity <dbl>, orgin_lon <dbl>, orgin_lat <dbl>,
## #   destination_lon <dbl>, destination_lat <dbl>

destPlot()

destPlot() takes a start and end city (from the cities in fd_heatmap_group and returns a ggplot map object with a line between the two cities plotted over a map of the United States. This function uses flightUi2() to get the start and end coordinates of the two cities, and then uses those coordinates to define a line and two points on the a ggplot object. This function what creates the dynamic map in the app.R UI.

#initialize ggplot base objects for function to work
world <- ne_countries(scale = "medium", returnclass = "sf", country = "united states of america")

map <- ggplot(world) + geom_sf() + coord_sf(xlim = c(-126, -66), ylim = c(25, 49), expand = FALSE) + theme_nothing() 
destPlot <- function(start, end){
  subset <- flightUi2(start, end)
  return(map + geom_segment(
  aes(x=orgin_lat, y=orgin_lon, 
      xend=destination_lat, yend=destination_lon),
  show.legend = FALSE, 
  data = subset, color = "blue") +
    geom_point(aes(orgin_lat, orgin_lon), data = subset) + 
    geom_text(aes(orgin_lat, orgin_lon, label=OriginCityName), data = subset) +
    geom_point(aes(destination_lat, destination_lon), data = subset) + 
    geom_text(aes(destination_lat, destination_lon, label=DestCityName), data = subset)
)
}

destPlot("Minneapolis, MN","Denver, CO")

Creating Visualizations

ggplot Object Setup

world <- ne_countries(scale = "medium", returnclass = "sf", country = "united states of america")

map <- ggplot(world) + geom_sf() + coord_sf(xlim = c(-126, -66), ylim = c(25, 49), expand = FALSE) + theme_nothing() 

This first map is of the flights with the greatest avgDelay or average delay in minutes.

map + geom_segment(
  aes(x=orgin_lat, y=orgin_lon, 
      xend=destination_lat, yend=destination_lon,
      color=avgDelay, alpha = avgDelay),
  show.legend = FALSE, 
  data = fd_heatmap_group_continetnal
)

This last map is of the most popular flights in the US, the transparency of the line being determined by qauntity or the number of flights in a given group.

map + geom_segment(
  aes(x=orgin_lat, y=orgin_lon, 
      xend=destination_lat, yend=destination_lon, alpha = quantity),
  show.legend = FALSE, 
  data = fd_heatmap_group_continetnal, 
  color = "#418733"
)

About

Will my flight be delayed? Built with R, this app predicts the delay of a flight between any two US cities.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published