Pivotal Data Science Transport Demo

This demo predicts the duration of currently active road traffic incidents in London. Incidents are unscheduled disruptions caused by collisions, surface damage, burst water mains etc, and do not include planned roadworks or other scheduled disruptions.

Find out more on the Predictions, Models, and Technology pages.


In order to create the predictions for the duration it is necessary to understand more about the incidents and select the best machine learning algorithm for the task.

Unexpected Incidents by Date

First let's look at all the unexpected incidents recorded since September. Here we can see the number of active incidents every two hours since we started collecting the data. If you hover over a particular point you can see the time, number of incidents and the weather conditions at the time. The peak on 28th of October is due to the effects of a severe storm that hit the UK.

We can also look at all the traffic disruptions over this time period. Select the "All Disruptions" button to add these to the chart. Most of these other disruptions are scheduled and long running roadworks. At any one time there are a lot more active disruptions than just unexpected incidents. However the total number of active disruptions is normally only around double the total number of the generally shorter unexpected incidents.

Use the focus chart to zoom in on a particular date range.

Mean Number of Unexpected Incidents by Day of the Week

One of the easiest ways to understand the data is by considering how many unexpected incidents occur at different times of the day. Here we show the average number of incidents for each hour of the day across each day of the week.

Select and deselect which days to view using the legend buttons.

Comparison of Duration Times for different Subcategories

For the disruptions that we are looking at there are a variety of different subcategories, each with its own distribution of durations. Here we can compare these and gain some intuition into how long we might expect a "Burst Water Main" incident to last compared to a "Collision" for example.

In this plot the area under each curve is the same, but the distributions are peaked at different durations. Breakdowns and Collisions have the shortest incidents on average, and Surface Damage and Burst Water Mains experience a full range of durations which extend well beyond the edge of this graph. The average Burst Water Main incident lasts over 100 hours, and the longest one in the data so far was 641 hours or over 26 days in duration.

Select and deselect which subcategories to view using the legend buttons.

Comparison of Duration Times for Rainy or not Rainy Conditions

Weather conditions might be expected to heavily affect the number and duration of traffic incidents. From the graph above we can see that the number of incidents is indeed increased, with stormy conditions particularly in late October 2013 contributing to a massive increase in the number of incidents.

On the other hand the chart below show the effect on duration times of rainy conditions compared to when it is not raining. If you live in LA you might think that it rains in London all the time, and while that's not quite the case there are certainly a large number of rainy days. In this plot however we can see that incidents which start when it is raining are more likely to have a shorter duration than those which start in dry conditions. We have not taken into account the severity of the rain in this analysis but this result is perhaps somewhat counter-intuitive.

Location of incidents

This map shows the location of all recorded incidents. The intensity corresponds to the duration of the disruption(s) at that location with incidents lasting over 1 hour showing in red.

What's next?

Learn how the insights gained from this analysis be used to predict the duration of future incidents in the Models section.

Powered by:

Created by Ian Huston | Twitter | LinkedIn | Website

Thank you to the whole Data Science team for their help in producing this demo, especially Noelle and Vatsan.