<aside> <img src="/icons/info-alternate_blue.svg" alt="/icons/info-alternate_blue.svg" width="40px" />

Introduction

Originally a Hackathon challenge at HackTrain which my team had won. This project was a POC with ScotRail (Network Rail), and Atkins as the solution partner (my previous employer).

Severe weather conditions in Scotland is a major engineering challenge for the rail industry, as it causes faults, asset failures, leading to disruptions, delays and cancelled trains → massively affecting customer journeys and cost implication to the industry in millions.

The developed POC was attempting to address this engineering challenge by using ML to provide weather based asset failure predictions, which would allow to pro-actively minimise risks by adequate planning and maintenance.

</aside>

<aside> <img src="/icons/condense_yellow.svg" alt="/icons/condense_yellow.svg" width="40px" />

POC Solution

The developed POC was a collection of ensemble classification models developed using Catboost that predicted probability of each of the different types of asset failures for a given weather condition and recent weather history; which would be visualised on geo-spatial dashboard for a given data in near future based on the weather forecast.

</aside>

<aside> <img src="/icons/bullseye_green.svg" alt="/icons/bullseye_green.svg" width="40px" />

Project Outcome

The project was delivered as a solutionised POC, for Network Rail’s team for further work upon; but due to bad quality data labelling, the advice was to re-label and plan a further data collection phase before the POC can be considered viable.

</aside>