2
$\begingroup$

Im studying M.Sc Data Science and in the module "Decision Support Systems" me and my group have to make a presentation. Our Proposal is the following:

Background With generally high demand for runway access and complex organisation required to plan efficient flight connections, delays in airplane arrivals or departures can be costly to airlines and passengers both. While many of the causes of delay, such as weather, mechanical failure, service issues, or strikes can be tricky to predict far in advance, there may be patterns to delays that may be discovered based on arrival and departure statistics for prior flights, and used to predict whether a future flight will be on time.

Objective & Deliverables By the end of the project, we hope to have developed a model that will determine accurately (based on an AUC greater than random chance) whether a flight will be delayed, and by how much, based upon the variables present (airline, airport, flight time, etc.) which is independent of knowledge of weather predictions, or status of prior flights. Scope The scope of this project will be the creation of a model, using either Monte Carlo or Random Forest, for the purpose of predicting whether a flight will be delayed, based upon independent variables such as day of week, airline, airport, flight duration, etc. Dependent variables, such as predicted weather or prior flight status will not be included or considered for the purpose of the model.

Methodology Utilising Python code, we intend to follow a OSEMN Pipeline. First, obtaining the data from Kaggle. Second, import and scrub/pre-process the data to check for NAN values and add categorical data conversions (for Random Forest option). Third, following this we will undertake exploratory data analysis and create visualisations in order to identify key variables and trends. Fourth, Build a model and assign test and validation data sets, train the model. Fifth, interpret the results and compare accuracy using AUC to gauge the usefulness of the model and check for overtraining.

Dataset We will be using data provided by the U.S. Bureau of Transportation Statistics, covering a time period of 01 January 2019 to 06 December 2019, which tracks the on-time performance of U.S. domestic flights operated by large commercial air carriers. This data provides summary information on the number of on-time, delayed, cancelled, and diverted flights, along with cancellation and flight data. The raw dataset, compiled and posted by a user on Kaggle, contains 29 points of data for 484,521 flights.

https://www.kaggle.com/datasets/undersc0re/flight-delay-and-causes?resource=download

So, we want to build a prediction on Random Forest, Decision Tree and a third model (Monte Carlo Simulation if its suitable) and then compare all 3 models. My part is the MC Simulation but im struggling to find a start. Because I only found Stock Prediction on that using the Pandas_Datareader but I have no idea how to deal with it by using a (Kaggle) Dataset.

Do you guys have any recommendations in regard to the MC Simulation or maybe can recommend another model? Would be great! Thanks, Wayne

$\endgroup$

1 Answer 1

1
$\begingroup$

Weather is responsible for 90% of the flight delays. How is it possible to make reliable predictions with just 10% of the remaining causes? (if their data is available)

You have an existing map called Misery Map where you can see a strong correlation between bad weather and delays.

If you click on the "play" button, you will see that everytime there is rain above an airport, the delays increase significantly.

Without weather, the best you can do is dealing with high traffic hours (generally around 11am or 5pm): When the traffic is more dense, there are generally more delays.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.