Do the most “scenic” spots around London also experience fewer road traffic accidents?

Introduction

In this demonstration, we will generate a risk map of London roads using traffic data and juxtapose it with a dataset of scenic spots around the city to explore their correlation. While it's reasonable to assume that scenic spots are often situated away from busy main roads prone to accidents, here we aim to validate this assumption.

Open Fused Graph in conode

Datasets and upload

The datasets we will use to answer this question are the following

1. STATS19 STATS19 is the UK's public accident database, containing police reports of road traffic collisions that took place around the country. We have selected a subset of the dataset focussed on London incidents, which will provide us with a base risk-map.

2. TLF JamCams To build a thorough risk map for London roads, we can't just rely on STATS19 data as the records only cover police-reported incidents resulting in injury. To consider the near-miss events which also impact road risk, we scraped jamcam clips—public live streams from London's busiest intersections

3. ScenicOrNot Dataset The ScenicOrNot dataset contains images of locations around the UK with a "scenicness" rating, assigned by the public (you can vote here!)

Each of these datasets comes in the form of a .csv file. STATS19 contains crucial insights on the risk associated with UK roads, however it is a complex dataset with over 50 features making manual querying overwhelming. While STATS19 provides links to images and Jamcams to videos, navigating them through the tabular format is less than ideal.

Let's integrate these three separate datasets into one interactive fused knowledge graph in just a few minutes.

Building a risk map from STATS19

Focussing first on STATS19, upon upload the dataset is automatically converted into a knowledge graph which is then presented back to you in the form of views such as spatial views, embeddings, and histograms.

In the "Spatial View" we can immediately observe where all the road traffic collisions took place.

As the knowledge graph is fully interactive, you can select any node on the spatial view that represents a traffic event, and read through all the features it contained using the Inspect panel on the right.

Additionally, we can examine the URLs automatically attached by the bolstering agent to each data point, which in this case are Google Maps links. Selecting any event node and hitting the shortcut u (or navigating from the Node -> Open URL) allows us to explore the local environment where collisions occurred, providing valuable context and real-world validation.

empty line

empty line

To build a simple risk map, let's select one of the features from the "Home" view which indicates the severity of each event, such as number_of_casualties, and represent it as the node size in our spatial view. Additionally, we'll assign all these events a red color to distinguish them from the next dataset we're about to import.

empty line

Fusing in JamCam videos

Now let's incorporate the JamCams dataset. If uploading directly, a few views will open as suggested by conode, however as we are only interested in the spatial view we will close all others. Before copying and pasting the JamCam nodes into the STATS19 Spatial view, let's colour the nodes representing JamCam videos black and enlarge their size, to visually differentiate them from STATS19 events.

We can now interact with both the STATS19 dataset and JamCam videos from one view in conode! Click on the Jamcam nodes and use Watch Video (accessible through the Node menu) to check out the space of near miss events our risk map now considers.

empty line

Fusing in ScenicOrNot images

Now that we've built a quick risk map of London's roads, let's add in our scenicness profile. Upload, colour, then copy and paste the scenicness nodes into the spatial map, or click on the following image to access the fused spatial view.

Here is an example of a highly scenic location, which we notice is located in an area of very few traffic incidents

Dataset Comparison and overview

We can now navigate around the fused view and observe that the regions of high-risk traffic collisions often take place in areas with new highly-rated scenic spots, and vice versa. As expected, the most scenic locations in London do not correlate with spots at high-risk of traffic collisions.

In summary, we've showcased how you can leverage conode to: merge multiple CSVs representing different types of media, interact with them seamlessly in a unified platform, and promptly address queries involving all data sources, all accomplished in just a few minutes!