Data-Driven Holiday

I knew we would arrive in St. Pancras International Station and stay somewhere accessible by transit.

So I pulled location data for every London Underground, Overground, and National Rail station.

Then I pulled data about each train line.

I found the stations that were accessible from St. Pancras within one hour, estimated by Transport for London's journey planner.

Then I estimated a 10-minute walkshed around each one.

To estimate a ten-minute walk, I used a straight-line, 800-meter buffer around each station – a standard shortcut I've seen while working on my master's thesis.

Once I had the walksheds, it was time to find hotels.

I used a web scraper to get a list of hotels within the walksheds under 120 euros per night.

This list was joined to a list of accommodations including geographic coordinates.

Then I filtered to hotels above 60 euros per night, above a 3.8-star rating, and located within the walksheds.

This narrowed down the list to 430 candidate hotels, spread across 227 stations.

This was a good start, but not all of these hotels were in equal locations. Some were in boring suburbia, others in rough parts of town, and others by the airport.

Hardly a good first impression.

So the next step was to analyze the quality of the area around each station.

And this is where I went a little overboard.

I started by brainstorming a list of things that make a place feel pleasant – lots of street trees, parks nearby, and cultural heritage.

Then I made a list of things that make a place feel unpleasant – gray asphalt, traffic, and noise.

To quantify these factors, I pulled environmental data, greenery data, architecture and infrastructure data.

I clipped each data layer to every walkshed and measured how much of the station area was covered by trees and open green space.

Then I counted how many nationally designated historic structures fell inside each walkshed.

Then I estimated road widths and combined streets and parking lots into one measure of land taken up by asphalt.

The noise data was a grid of 10-meter cells, so I counted the share of cells at or above 55 decibels.

Now I had a bunch of numbers for each walkshed, but I didn't know how to weigh them.

More trees is probably good – but how much does noise matter? What about asphalt?

Luckily, I found an empirical way to check.

Scenic-Or-Not is a website where visitors rate photos of places across Britain on how attractive they look – over 200,000 photos, each rated by at least three people.

In 2017, a research group at the University of Warwick trained a machine learning model on the Scenic-Or-Not ratings and used it to predict scenic ratings for thousands of geolocated photos in London.

I matched their predicted scenic ratings to my walksheds and tested which of my metrics tracked them.

Several metrics showed statistically meaningful correlation across all 264 walksheds. For example, tree cover and open green space predicted higher scenic ratings. Noise and asphalt predicted lower scenic ratings.

I combined these and multiple other factors into a scenic score for each walkshed.

Then I compared my model to the one produced by the research group based on the average Scenic-Or-Not ranking inside each walkshed.

But the Scenic-Or-Not model did not show any correlation with something I consider important: architectural built heritage.

So I kept listed structures as a separate heritage score for architectural character.

I now had two scores for each walkshed: scenic and built heritage.

I played with various weights to see what happened if I favored heritage more or if I filtered to those within a kilometer of a listed park, garden, or cemetery.

I tuned the dials to something that I found appealing. In my opinion, the best candidates clustered around two places in London.

The neighborhoods around Kensington Gardens – Queensway, Holland Park, Bayswater, and Notting Hill Gate – hold some of London's celebrated Georgian townhouses.

The area is accessible to transit and amenities with many hotels.

And the others clustered around Greenwich Park: Greenwich, Cutty Sark, and Blackheath.

From among the available hotels in both areas, we chose one in Blackheath: The Clarendon Hotel.

It's accessible in about 40 minutes by train from St. Pancras station, close to a charming village high street, the Royal Observatory in Greenwich, plus several museums, parks, and scenic lookout points.

And as a bonus for a geographer, it's only 1.28 km from the official Prime Meridian marker.

Sold.