Case StudiesCorrectional Bed Demand Forecasting
2.18% - Mean absolute percentage error across a 63-month validation window.
Challenge
2.18% mean absolute percentage error across a 63-month validation window. Methodology published as a working paper.
What we built
A nonlinear autoregressive neural network with exogenous inputs, trained on admissions and remand flow with upstream justice system volumes as driving variables. Validated across a five-year out-of-sample window rather than fitted to it - the forecasts were tested against what actually happened, not optimised to match what was already known.
The model
Outcome
Methodology published as a working paper.
Client details and delivery specifics are confidential. References available on request.
Ready to learn more?
Making capital or capacity decisions on forecasts you don't fully trust?
Let's talk about your problem