London Property Finder, applied machine learning

Role: Applied machine learning, self-directed research. Timeline: 2026.

An interactive tool that scores live London property listings against a gradient-boosted machine-learning model, built to test whether publicly available data alone can identify mispriced homes.

The question

Most house-price models predict a number and stop. The question here is sharper: using only publicly available data, can a model flag which current listings look mispriced relative to comparable sold properties, and can the resulting tool be interrogated by the user rather than trusted blindly?

Approach

• Training data. 2.18 million Land Registry transactions spanning 1995 to 2026, deflated to a common reference with the House Price Index.
• Model. A single gradient-boosted regressor (XGBoost, Optuna-tuned) over 29 features drawn only from public sources — listing attributes, ONS area statistics, and spatial signals from nearby sold prices. An earlier stacked version was deliberately rebuilt to remove a subtle asking-price data leak, trading a flattering headline number for an honest, sold-price-only model whose mispricing signal is real rather than circular.
• Tool. A single-page interface scoring roughly 82,000 live London listings, with sold-nearby comparables drawn from the Land Registry, filter and map views, multi-key sorting and CSV / XLSX export.

Result

• ~50% of predictions land within ±10% of the actual sold price on held-out data, with a median absolute error under 10% and a mean absolute error of about £96,000 — measured with no asking-price leakage.
• On a fresh scrape of ~90,000 live Rightmove listings (enriched through a leakage-free pipeline), a median absolute gap of 15.9% between the model’s valuation and the asking price — tighter than the earlier leaked model, despite using no asking-price information at all.
• The headline accuracy is lower than the earlier leaked version by design: removing the leak is what makes the numbers honest and the mispricing signal trustworthy.

Stack: Python, XGBoost, Optuna, geospatial analysis, leakage-free feature engineering and a Land-Registry comparables engine.

Open the live finder · try the single-property valuation tool