| Student's name | SCIPER |
|---|---|
| Nicolas Karmolinski | 316655 |
| Roméo Maignal | 360568 |
| Jakub Kielar | 423372 |
The primary dataset we will be working with is a dataset of plane crashes from 1908 to 2024 found on Kaggle by Luiscé Francisco. This was the most up-to-date dataset we could find that contains relatively complete information on crucial details for our analyses: route, operator and aircraft type. The dataset contains information on every kind of flight possible, thus will require some processing to only keep commercial flights, which is our project's main point of interest. Some slight cleaning might be necessary. The dataset contains many columns, including a couple that we will likely not need, like fuselage number or flight number, the latter of which is quite incomplete. The route column may need to be split into two departure and arrival columns.
As a complement, we will be using the airports, routes, and airlines datasets from OpenFlights, a free, high-quality and comprehensive aviation database. In conjunction with our primary dataset, we will be able to relate the plane crashes to geographical locations, priming us for apt visualizations.
No matter how safe air travel might be, it's always difficult to completely shake off the feeling of insecurity of flying.
In an attempt to address these concerns, we aim to provide an interactive platform to assess the safety of air travel and the variable risks associated with specific criteria such as routes, airlines, and aircraft types through various visualizations, providing visitors with insightful analyses on the reliability of agents involved in air travel. We also intend to pursue some more in-depth analyses, comparing advanced statistical results across different actors in the aviation industry, like analyzing the frequency of accidents through the lens of a Poisson distribution.
Some basic cleaning and preprocessing to the crashes dataset is done in preprocessing.ipynb. The main operations aim to simplify the dataset by dropping unnecessary columns from the original and formatting columns such that we can relate them back to the OpenFlights datasets. Regarding the latter, as they are high quality, no preprocessing was deemed to be required.
For our final exposition, we aim to address our research objectives through advanced statistics and provide an insighful interpretation of our results. In the meantime, in order to give a preliminary overview of our project, we explore a few fields within our datasets in the following notebook: preliminary_analysis.ipynb .
The exact dataset we used is technically not yet widely described by researchers. However, a similar dataset, closely related to one we will use, has been widely investigated and described. We didn’t choose that one is due to the fact that its only tracks records until 2009, which makes it slightly outdated for use in modern solutions.
That dataset was investigated and visualized in Airplane Crashes Data Visualization, presenting a lot of important information with plots and classification graphs. It investigated the fatalities, differences between military and civil flights statistics, the reliability of operators and many other factors.
Further investigation of the dataset was found on Kaggle: Who not to fly with, focusing on the accidents and fatalities connected to flight operators. It aims to present reliability of each operator throughout the history.
Another article presents crash analytics for models of planes, manufacturers, crash locations and even comparison between early and late hours of the day.
Ultimately, we want to present an interactive map of the world with routes of flights and places of crashes.
One of the first inspirations for our work and presenting the routes of given flights was the popular FlightRadar, where the flight can be traced by its unique flight number. The inspiration for presenting data as a map was also the previously mentioned OpenFlights website, where we found their map of flight connections, but in a very static manner. We wanted to go a bit further and make the map interactive, enabling users to interact and pick single connections between airports and see the crashes of that very route. This is led us to FlightConnections, which presents this data in a much more organised and interactive way, where users can pick starting points to see the possible route choices. We wanted to follow such an approach, but with the presentation of additional data on crashes, a combination that we didn’t encounter during our investigation.