Prices come from an XGBoost regression model trained on 114,594 Sold transactions (2005–2026) using a time-based 70/15/15 split, with locked hyperparameters from the ML report.
The model uses 34 features: property attributes (beds, baths, land size), location (suburb, distance to CBD, train), and area demographics (income, age, population, crime). At inference, current Year and Month are injected so predictions reflect today's market level.
Deal Signal logic: compares asking vs predicted price. Asking >10% below predicted = Good Deal, asking >10% above predicted = Overpriced, otherwise Fair.
The map has two modes - toggle them using the buttons at the top right of the map panel.
Fill in the property details below, click Estimate Price, and the XGBoost model runs directly in your browser (no server) to return a point estimate plus an 80% prediction interval.
The model is loaded once per session (~14 MB total across 3 quantile models). Predictions are calibrated against the latest weekly snapshot and reflect the current market level.
The model will load on first use (~14 MB), then runs instantly in your browser.