Accurately predicting plane ticket prices for an airline

Customer

A leading short-haul airline operating in more than 30 countries.

Problem

The airline operates in a highly competitive market. Hundreds of thousands of passengers fly on hundreds of routes to multiple destinations in Europe every day. With huge choice, price is the critical differentiator. Prices for flights fluctuate daily, even hourly, so it is crucial for the airline that it knows how much its competitors charge. The airline asked Faculty to build a model that could predict its competitors’ prices in the next three months, so that it could price its own flights accordingly.

Solution

First, we analysed price data on six different routes by three competitor airlines over the past few years. This analysis showed that prices were volatile over the 90-day period before a flight’s departure. However, there were some common patterns, including identifying ‘take-off’ prices (prices that stayed low until nearer take-off time, when they increased, ‘false landing prices’ (prices that started off high, dropped and then reverted to high) and ‘turbulent prices’ (which were volatile).

We then built a model (XGBoost) that could predict the ticket price of a given airline for any day up to 90 days in advance. The predictions generated from the test data set were compared to the actual cost of tickets by the airlines on those days.

Impact

Our model was able to generate predictions that were between 70% and 80% accurate up to 90 days before each given flight.

The airline took this foundational data science work to expand its price prediction model, which it intends to build into its pricing strategy across Europe.