We were delighted to welcome Auto Trader to Alan Turing last Wednesday for an industry problem solving event focused on data analysis.

If you’ve ever looked at buying a car online (or, like me, enjoy a bit of fantasy car shopping) you’ll doubtless have come across Auto Trader’s website; it’s the UK’s largest digital automotive marketplace with around 48 million visits a month. Combine this site traffic with a plethora of search filters and you get a huge amount of data to deal with. What can maths tell us about users’ search habits and the cars they’re buying?

Cue Auto Trader data scientist Dr Peter Appleby, who presented us with two problems that reflect the mathematical challenges the company faces, each focusing on a different technique for data analysis. The first was using regression on historic car prices and mileage to create a model for valuation. The second looked at applying clustering to try and understand the search space generated by users using the different filters- namely, which combination of filters (make, model, fuel etc.) are more common than others?

After a quick crash course (pun intended) on regression and clustering, we headed over to the computer cluster to wrangle with some real-life datasets. Of course, there’s only so much you can do in an hour, but there was plenty of discussion and coding going on as people tried to extract useful insights from the data. The clustering problem proved most popular -the conclusion from the session seemed to be that users who search for Audis also tend to look for BMWs and other pricier makes. Not exactly the most ground breaking conclusion you might say, but one can imagine extending this method to a complex multidimensional search space, where much more interesting patterns might start to emerge.

In all, it was a fun day for both those familiar with these techniques and those who were trying out data analysis for the first time. As one undergraduate attendee told me over coffee at the close of the event, if didn’t matter that he wasn’t familiar with the techniques beforehand “it’s fun just to come along and have a go, mess around with different models- you’re learning something new.” For the more experienced, it was “motivating to get to work with a real dataset, rather than making up your own.”

For my part, I wouldn’t have necessarily realised the extent to which maths is used behind the scenes at somewhere like Auto Trader – it’s really interesting to see how maths and data-driven methods are becoming an increasingly important tool for companies in the digital age. I would like to offer my personal thanks to Dr. Appleby and Auto Trader for their support for this event, and I look forward to organising the next industry problem solving event!