The original idea I had for the final project was to create a complex app for On Century Avenue. In fact, the idea was to create a sort of Century Media app, as i got several rumours about OCA and Feast (the NYUSH art zine) being merged. However, the project for the organisational/logistic reasons would most likely remain unused. Firstly, the decision turned out to be that OCA and Feast will become separate media outlets, but on the same server. Secondly, the current pages are not using WordPress and rely heavily on editorial/graphic work of people who have no experience with coding whatsoever, so even despite the fact it’s fairly simple to implement a Reactive front end to WordPress API, it will definitely have an extra cost in terms of having the entire editorial team go through all changes in content creation settings.
Therefore, I decided to go for the idea that was second on my priority list, but might be interesting to explore.
My idea is to create a database and search engine for future flights within the United States, with the functionality of flight prediction.
The main question the app would answer would be: is there a reasonable possibility that my flight will be delayed at a given time, day, place and airline?
Last semester in Amsterdam, I studied machine learning. Together with Dorian Buijse, my fellow classmate from the Amsterdam Uni College campus, we developed a prediction model of flight delays within major airlines and airports in the US. Our dataset consisted of all US domestic flights from year 2015, but we filtered it very thoroughly due to a lack of computing power and time.
The features we used for prediction were:
– Month (numeric)
– Day of the week (numeric)
– Departure time (numeric, converted to full hours)
– Airline (converted to dummy variables)
– Departure Airport (converted to dummy variables)
– Arrival Airport (converted to dummy variables)
Using a multi-layer perceptron neural network, we achieved an accuracy of 65%. Given the complexity and difficulty of extracting relevant features from the set, it was a relatively decent score, but I’m sure it can be improved if extra features are added. This gives a possibility for an app that would predict whether or not a flight delay is likely.
A very well-designed, clean (Material-designed of course) and distraction-free way to find information about current flights and find bookings for the future.
At first, user is able to choose a country where they reside/are flying from.
Then, user is presented with a flight search screen. Looks very familiar, works basically the same way as SkyScanner, Kayak or most airline pages. You are able to choose between roundtrip and one-way options, you have two calendar pickers.
Below the search component, user can find popular destinations from a list, or click them from a Google map.
Once user makes a search, they can see the flight combinations, price-ascendingly.
But there is another Flights feature within the Google search engine itself. If a user searches a flight number, they are able to see all of the info about flight departures, gates and officially announced delays. But a feature that’s even more interesting is the delay prediction notice. Apparently, when a flight is likely to be delayed by at least 30 minutes, a notification about a potential delay shows up.
My app is not going to be an all-in-one app for flights, unlike Google’s.
Instead, my main focus will be the flight delay prediction. I want the user to be able to predict a potential delay given the data available about a future flight, as well as to compare the airlines across the punctuality criterion.
The sample user stories will be:
User, after entering the app, can type in a location of the departure airport using the city name or airport code.
User, after picking a departure city, can choose the date and time of departure.
User, after picking the departure city, date and time, can compare different airlines in terms of their likelihood of delay on that day.
User, after picking city, date, time and airline, can see whether or not a flight is likely to be delayed.
week 1. Wireframing. Tweaking the ML algorithm, getting the Python-based API to output data
week 2. Connecting the API with a React app. Writing the code
week 3. Writing the code, final tweaks, user testing