A UK Data Science blog.

UK House Price prediction model overview

Models

Active

Model 1.9 – Our forecasts come from a hierarchical machine-learning model that works down through three nested tiers — national, regional and local — with each layer feeding into the one below it. A single unified model covers every horizon from six to sixty months, trained on roughly two decades of data (2006–2024) and drawing on a broad mix of macroeconomic, labour-market, housing-supply and price-momentum signals. It’s validated strictly out of sample on 2024–2026 data it never saw during training. As you’d expect, accuracy is strongest at the national level and over shorter horizons, with error rising gradually the further out — and the more granular — the forecast becomes. Reassuringly, the gap between training and test error is small (around 1.16×), which suggests the model is generalising to genuinely new conditions rather than simply memorising the past.

No Longer in use

• Model 1.8
• Model 1.7
• Model 1.6
• Model 1.5
• Model 1.4
• Model 1.3
• Model 1.2
• Model 1.1