(Image Credit: iStockPhoto/iLexx)

Hotel cancellations are a big concern for the travel industry as rooms are held for people who may not even turn up at not just a significant cost to the business, but also preventing other travellers from using the room which may be required for their visit. 

Nigerian hotel booking service, Hotels.ng, held a hackathon on August 1^st which aimed to answer whether it is possible to predict what users are going to cancel a hotel booking through use of machine-learning. After all, even the slightest improvement on reducing cancellations has the potential to have a significant economic impact. 

In a press release announcing the event, the organisers wrote: "The idea behind the hackathon is simple: all hotel booking websites around the world have incidents of cancelled hotel bookings. Some users are more likely to cancel their bookings than others, depending on several factors/conditions. For example, a person who books and pays is less likely to cancel his booking compared to someone who intends to pay on arrival. A person who books 5 different hotels in different states on the same day is likely to cancel at least 4 of those reservations. Or maybe not – because he’s probably booking for other people." 

Around 30 people attended the event and hacked throughout the day using a dataset of 5,000 pieces of mock data to create an intelligent algorithm to predict which users are likely to cancel their booking – with incredible results. One team accurately predicted 50% of all the people that would cancel, and a second team predicted 33% of the cancellations. 

The winning team, Andela, said their first step in creating their algorithm was identifying key attributes for a particular booking that are most likely to influence a cancellation. After determining these, Andela says they derived quantitative weights based off the appropriate class an attribute belongs to and then applied decision tree learning algorithms to derive their final predictions. 

Machine-learning is an incredible science which can have a significant impact across a variety of industries, as this hackathon goes to show. We look forward to seeing how similar hackathons will harness machine-learning techniques to create innovative solutions for industry-wide problems. 

Do you think more hackathons should focus on machine-learning solutions? Let us know in the comments.