Skip to main content

Flood Alert - Part 9: Project Reboot with Machine Learning

Future Hopes for Flood Alert V2.0 - with Machine Learning

Above: Original Video from initial Pilot.

Early Sketch for Concept for Flood Alert

Above: Early Sketch for Concept of combining the API data with local user data/feedback.

A New Chapter...

This concluded this version of the project. It is not presently much more sophisticated than the Environmental Agency API, other than it is more prominent as an object, as opposed to an App, for those who are less attentive to smartphones/devices!

However, as mentioned, the intention of the project is to develop it further, so the ‘hyper-local’ aspect becomes more relevant to you / chosen users - as it will over time begin to ‘dial-in’ the specific factors which influence your area of interest. Perhaps your home, school, office, etc.

Having lived with the Alerts from the Environmental Agency (one going off even as I write this, today, 22 Sept at 3:37pm), I’m aware that these would benefit from me ‘voting’ if it’s a relevant alert for me. Personally, I’ve come to not be overly concerned with an  ‘Amber’ alert as there have been many ‘Amber’ alerts in the past that did not affect my area.so I’d likely ignore it, and not ‘downvote’ it to ‘Low Risk’.

To manage all of these ‘local calibrations’, this can be done with basic ‘look-up’ tables where the data is more straightforward, but I do think that we are starting to have a meaningful case for the addition of AI to help. As much as I think there is a lot of hype and over-zealous claims around AI, this seems to be a genuine use for a mix of supervised and unsupervised learning.

On a supervised level, we are telling the AI ‘yes’ or ‘no’ on its ‘guess’ derived from the Environmental Agency API. However, where it gets interesting is to consider how this ‘vote’ also collects some supplementary data, in the form of a snapshot of what the weather is like and has been like the last week for example. This is (I think) best defined as ‘unsupervised’ learning, and we’re asking the AI to look at ALL the data available and spot patterns, but still verify with the Supervised data and Vote on what the user thinks is most likely.

I’ll be honest that at this point I’ll be needing to work with some folks who are a little more skilled at the AI/ML work, and indeed have a more statistical background to understand confidence intervals, probability, error, etc. So this will be a learning curve for me, but from asking around so far, it seems a fairly intriguing project for a few AI/ML folks to work with - being neither too complex, but also making a useful feature of user input, or ‘Citizen Science’ as an output of the project.

Please get in touch if interested in collaborating on this exciting next stage of the Flood Alert project...

Contact: https://www.judepullen.com/

mceclip0_c5b2ecdcd604598d3d6ad2d6ad8185d9ec1deff3.png

Blog Series Contents:

Prologue - The Case for 'Hyper-Localisation' of Civic Data

Research & Development:

Part 1: Filling the Local Data Gap
Part 2: Civic Services & User Experience Research
Part 3: Ideation of Flood Alert Concept
Part 4: Prototyping Back-Story
Part 5: Citizen Science Learnings

Open Source Build Guide:

Part 6: Build Guide for 3D Printed Assembly
Part 7: DIY Decals for 3D Prints
Part 8: Code & Data Guide

Future Ambitions:

Part 9: Project Reboot with Machine Learning

Winner of the 2020 Alastair Graham-Bryce "Imagineering" Award (IMechE), Jude thrives in high risk collaborations, uncertainty and pressure - drawing from global networks and experiences to deliver high profile campaigns and digital/physical products. A leading Creative Technologist & Physical Prototyping Expert, Jude has worked for NHS, Dyson, LEGO, and a number of start-ups. He is one of the eight featured inventors in BBC Two's Big Life Fix. More at: https://www.judepullen.com/

Comments