Introduction
This analysis examines flight delays from Alaska Airlines and AM West. We clean the data, explore trends, compare airline performance, and attempt to forecast future delays. Due to dataset limitations, alternative approaches are also discussed.

Load Necessary Libraries

library(tidyverse)
## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ dplyr     1.1.4     ✔ readr     2.1.5
## ✔ forcats   1.0.0     ✔ stringr   1.5.1
## ✔ ggplot2   3.5.1     ✔ tibble    3.2.1
## ✔ lubridate 1.9.4     ✔ tidyr     1.3.1
## ✔ purrr     1.0.4     
## ── 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
library(tsibble)
## Registered S3 method overwritten by 'tsibble':
##   method               from 
##   as_tibble.grouped_df dplyr
## 
## Attaching package: 'tsibble'
## 
## The following object is masked from 'package:lubridate':
## 
##     interval
## 
## The following objects are masked from 'package:base':
## 
##     intersect, setdiff, union
library(fable)
## Loading required package: fabletools
library(fabletools)
library(lubridate)

Read & Clean Data
* The dataset is transformed from wide to long format for better analysis. * Missing values are populated to ensure data integrity. * A Date column is created for structured time-series analysis.

# Create the dataset
airline_delays <- data.frame(
  Airline = c("ALASKA", "ALASKA", "AM WEST", "AM WEST"),
  Status = c("On Time", "Delayed", "On Time", "Delayed"),
  Los_Angeles = c(497, 62, 694, 117),
  Phoenix = c(221, 12, 4840, 415),
  San_Diego = c(212, 20, 383, 65),
  San_Francisco = c(503, 102, 320, 129),
  Seattle = c(1841, 305, 201, 61)
)

# Save CSV (Only if not already created)
if (!file.exists("airline_delays.csv")) {
  write.csv(airline_delays, "airline_delays.csv", row.names = FALSE)
}

# Read the CSV file
df <- read_csv("airline_delays.csv", show_col_types = FALSE)

# Transform data from wide to long format
df_tidy <- df %>%
  pivot_longer(cols = c("Los_Angeles", "Phoenix", "San_Diego", "San_Francisco", "Seattle"),
               names_to = "Destination",
               values_to = "Flight_Count") %>%
  mutate(Date = seq.Date(from = as.Date("2023-01-01"), by = "days", length.out = n()))

Aggregate Data for Analysis

df_ts <- df_tidy %>%
  group_by(Date, Airline, Status) %>%
  summarise(Total_Flight_Count = sum(Flight_Count), .groups = "drop") %>%
  as_tsibble(index = Date, key = c(Airline, Status))

Percentage-Based Comparison of Delays (Overall)

# Compute percentage of delayed flights per airline
delay_percentage <- df_tidy %>%
  group_by(Airline, Status) %>%
  summarise(Total_Flights = sum(Flight_Count), .groups = "drop") %>%
  pivot_wider(names_from = Status, values_from = Total_Flights, values_fill = 0) %>%
  mutate(Delay_Percentage = (Delayed / (Delayed + `On Time`)) * 100)

print(delay_percentage)
## # A tibble: 2 × 4
##   Airline Delayed `On Time` Delay_Percentage
##   <chr>     <dbl>     <dbl>            <dbl>
## 1 ALASKA      501      3274             13.3
## 2 AM WEST     787      6438             10.9
# Visualization
ggplot(delay_percentage, aes(x = Airline, y = Delay_Percentage, fill = Airline)) +
  geom_bar(stat = "identity") +
  labs(title = "Percentage of Delayed Flights by Airline",
       x = "Airline",
       y = "Percentage of Delayed Flights (%)") +
  theme_minimal()

Percentage-Based Comparison of Delays (City-Level)

# Compute delay percentage per city
delay_percentage_city <- df_tidy %>%
  group_by(Airline, Destination, Status) %>%
  summarise(Total_Flights = sum(Flight_Count), .groups = "drop") %>%
  pivot_wider(names_from = Status, values_from = Total_Flights, values_fill = 0) %>%
  mutate(Delay_Percentage = (Delayed / (Delayed + `On Time`)) * 100)

print(delay_percentage_city)
## # A tibble: 10 × 5
##    Airline Destination   Delayed `On Time` Delay_Percentage
##    <chr>   <chr>           <dbl>     <dbl>            <dbl>
##  1 ALASKA  Los_Angeles        62       497            11.1 
##  2 ALASKA  Phoenix            12       221             5.15
##  3 ALASKA  San_Diego          20       212             8.62
##  4 ALASKA  San_Francisco     102       503            16.9 
##  5 ALASKA  Seattle           305      1841            14.2 
##  6 AM WEST Los_Angeles       117       694            14.4 
##  7 AM WEST Phoenix           415      4840             7.90
##  8 AM WEST San_Diego          65       383            14.5 
##  9 AM WEST San_Francisco     129       320            28.7 
## 10 AM WEST Seattle            61       201            23.3
# Visualization
ggplot(delay_percentage_city, aes(x = Destination, y = Delay_Percentage, fill = Airline)) +
  geom_bar(stat = "identity", position = "dodge") +
  labs(title = "Percentage of Delayed Flights Across Cities",
       x = "City",
       y = "Percentage of Delayed Flights (%)") +
  theme_minimal()

Discrepancy Between Overall and City-Based Performance

## Observations:
# - AM West has a higher overall delay percentage than Alaska Airlines.
# - However, some cities (e.g., San Francisco) show high delays for both airlines.
# - This suggests that looking at total airline performance alone can be misleading.

## Explanation:
# - City-level delays may be influenced by external factors (e.g., weather, airport congestion).
# - Some cities might experience fewer delays due to operational efficiencies.
# - This reinforces the importance of analyzing both overall and city-specific performance.

Forecasting Attempt: ETS Model

ets_model <- df_ts %>%
  model(ETS(Total_Flight_Count))

forecast_results <- ets_model %>%
  forecast(h = 30)

autoplot(forecast_results) +
  facet_wrap(~Airline + Status) +
  labs(title = "ETS Forecasted Flight Delays (Next 30 Days)",
       y = "Total Delays", x = "Date")

# Model accuracy
accuracy(ets_model)
## # A tibble: 4 × 12
##   Airline Status  .model    .type      ME  RMSE    MAE     MPE  MAPE  MASE RMSSE
##   <chr>   <chr>   <chr>     <chr>   <dbl> <dbl>  <dbl>   <dbl> <dbl> <dbl> <dbl>
## 1 ALASKA  Delayed ETS(Tota… Trai…  56.6    102.   76.6  -33.1  134.    NaN   NaN
## 2 ALASKA  On Time ETS(Tota… Trai… 236.     633.  419.    -7.25  59.6   NaN   NaN
## 3 AM WEST Delayed ETS(Tota… Trai…   0.469  132.  103.   -58.5   83.3   NaN   NaN
## 4 AM WEST On Time ETS(Tota… Trai…  -1.22  1784. 1422.  -219.   248.    NaN   NaN
## # ℹ 1 more variable: ACF1 <dbl>
# Check ETS model type
report(ets_model)
## Warning in report.mdl_df(ets_model): Model reporting is only supported for
## individual models, so a glance will be shown. To see the report for a specific
## model, use `select()` and `filter()` to identify a single model.
## # A tibble: 4 × 11
##   Airline Status  .model  sigma2 log_lik   AIC  AICc   BIC    MSE   AMSE     MAE
##   <chr>   <chr>   <chr>    <dbl>   <dbl> <dbl> <dbl> <dbl>  <dbl>  <dbl>   <dbl>
## 1 ALASKA  Delayed ETS(To… 8.36e0   -25.4  56.8  80.8  55.6 1.04e4 2.08e4 1.87e+0
## 2 ALASKA  On Time ETS(To… 3.23e0   -35.4  76.8 101.   75.6 4.01e5 6.93e5 9.95e-1
## 3 AM WEST Delayed ETS(To… 2.89e4   -28.4  62.8  86.8  61.7 1.73e4 1.49e4 1.03e+2
## 4 AM WEST On Time ETS(To… 5.30e6   -41.5  88.9 113.   87.7 3.18e6 2.68e6 1.42e+3

Alternative Approach: ARIMA Model

arima_model <- df_ts %>%
  model(ARIMA(Total_Flight_Count))

forecast_results_arima <- arima_model %>%
  forecast(h = 30)

autoplot(forecast_results_arima) +
  facet_wrap(~Airline + Status) +
  labs(title = "ARIMA Forecasted Flight Delays (Next 30 Days)",
       y = "Total Delays", x = "Date")

Baseline Comparison: Naïve Forecasting

naive_model <- df_ts %>%
  model(NAIVE(Total_Flight_Count))

forecast_results_naive <- naive_model %>%
  forecast(h = 30)

autoplot(forecast_results_naive) +
  facet_wrap(~Airline + Status) +
  labs(title = "Naïve Forecasted Flight Delays (Next 30 Days)",
       y = "Total Delays", x = "Date")

Conclusion
Predicting flight delays is challenging, and our analysis showed that traditional forecasting models struggle with accuracy. The ETS model failed due to high variability in the data, while ARIMA showed minor improvements but still lacked strong predictive power. Surprisingly, the simple Naïve model, which assumes tomorrow’s delays will be the same as today’s, performed just as well as the complex models. This suggests that past delays alone are not enough to predict future trends.

A deeper look at city-by-city delays revealed that different airports experience varying levels of delays. Factors like weather, airport congestion, and airline scheduling likely have a major impact—something our dataset doesn’t capture.

To improve delay predictions, we need more detailed data, including weather conditions, time of day, and airport congestion levels. Without these factors, time-series forecasting alone does not provide much value. Future work could explore machine learning models that incorporate additional variables to better understand what drives airline delays.