Flight Research and Analytics Group

Authors: Annie Chan, Azucena Faus, Mohammad Mahmoudighaznavi

Company Industry: Air Transportation

Company Size: 3 (startup)

GitHub Repository

Tableau Flight Map Dashboard

Abstract:

The airline industry plays an important and critical role in the world’s transportation sector. Various factors such as weather conditions, air traffic control, airport operations, etc., have the capability to directly and indirectly affect airline services, leading to delayed and canceled flights. Understanding the effects of such issues in advance can prepare customers and airline operators for potential reasons for flight cancellations.

Problem Statement:

Customer Dissatisfaction with air travel is at an all time low. Most of this dissatisfaction has come from complaints about airfare, reservations, and cancellations, among others. Although travelers know the propensity for canceled flights during harsh weather conditions, it is still frustrating when it happens. If a customer can know in advance the likelihood they might experience a cancellationa while they are in the process of making a flight reservation, not only can they be mentally prepared, but airlines can choose to provide those flights at discount. This can therefore ease customers with regard to airfares, reservations and cancellations.

Goals:

The success of this project will contribute to the achievement of the following goals:

1. Combining weather and flight data to find connections and correlations between extreme weather conditions and canceled flights.

2. Training a model to predict cancellations on the flight dataset with weather condition data and predict whether a flight has a high probability of being canceled based on the weather for that date/time. We would like to attain an accuracy/F1 score/Sensitivity of: 0.8, 0.8, and 0.7, respectively.

3. Providing recommendations to airlines on customer service plans that should be offered to customers in the event of a flight cancellation.

Ultimately, the goal of this project is to provide customers with information regarding the likelihood of canceled flights due to weather, along with suggestions for which airlines would be best to go with, in case the likelihood is high. The airline recommendations would stem from customer satisfaction when they experienced canceled flights with those airlines, and the services they not only offer, but actually follow through with. This will not only help customers planning out their future trips in case of uncontrollable circumstances, but also improve the reputation of multiple airline companies.

Non-Goals:

On the other hand, there are several non-goals that the team is not intentionally solving with this project:

1. Forecasting weather - The project is seeking to find if we can adequately predict canceled flights based on weather conditions. Once this concept is proven, a weather forecasting model can be used in conjunction with the canceled flights predictive model.

2. Predicting airline performance - This project intends to find customer satisfaction relating to the added benefit provided by airlines in situations where a flight is canceled. It won’t provide predictions on the satisfaction level of the inflight services (seat comfort, on-board service, inflight entertainment, etc. ).

3. Predicting delayed flights - Flights given we have no reason code for delayed flights that might indicate they were delayed due to weather or other circumstances.

Data Sources:

One of the data sources comes from a relational database called Airline that contains US domestic flight information for January of 2016, along with information regarding canceled flights and whether those cancellations were due to weather. The second dataset was collected from Kaggle, which contains weather data for US airports for the years 2016-2021. This way, departure airport weather conditions can be added features for those flights that experienced cancellations due to weather. The final dataset contains more detailed daily weather conditions for singular airports of interest that we will be focusing on, such as JFK. Having more detailed weather data might help in predicting canceled flights.

Flight Data

Airport Weather Data

JFK airport Weather Detailed Data

Some risks with the data are that it is only for one month, so any predictions would have to be for the following year, in the same month as weather conditions are seasonal. More data would have to be collected to further prove this concept and provide more accurate predictions, especially if weather trends change over time (climate change).

The source data is stored in a public S3 bucket and will get there by way of SQL extraction of the necessary tables, and loading the comma-delimited files for weather:

s3://sagemaker-us-east-1-993410942383/content-project/airline_data/