← Back to selected work

Predictive analytics

Rethinking likelihood of failure

Questioned the use of Remaining Service Life as a proxy for pipe failure, researched Random Forest applications in Norway and Canada, and built an initial proof of concept using a synthetic dataset of approximately 30,000 pipes.

The question

Remaining Service Life is often used as a proxy for Likelihood of Failure. That proxy can mislead investment decisions when failure behaviour is sparse, non-linear and driven by factors a simple residual-life curve does not capture.

What was tested

International Random Forest applications were reviewed and an early proof of concept was built on synthetic data of approximately 30,000 pipes to explore whether a more defensible method could be framed.

Current status

Scottish Water's Analytics team is progressing implementation and testing with external expertise. The aim is to identify the most defensible method and the data it needs — not to claim a production model before the evidence supports one.

Evidence boundary: Synthetic data is cleaner than a real network and failures are sparse. Those limits are part of the research, not footnotes to it. No production performance metrics are claimed here.

Contact View CV