Model report card
Predictor accuracy — back-tested against actual prices
Comparison of past 30-day predictions against actual daily city averages. Refreshes nightly.
Awaiting first back-test cycle
How we measure
Methodology in one screen
We don't claim accuracy in the abstract — the model is back-tested every night against the actual prices it tried to predict.
What we compare
Each row in price_predictions has a city, fuel grade, prediction_date and predicted_price. We join it against fuel_price_logs for the same city/fuel/date and compute the absolute difference in cents-per-litre.
Mean absolute error
MAE is the average of those absolute differences. It tells you the typical "off-by" number of cents — both up and down — without letting positive and negative misses cancel each other out.
Window
Last 90 days only. The model recalibrates nightly, so older predictions reflect an earlier version and would skew the headline.
Minimum sample size
A city/fuel pair needs at least 3 back-test points to appear. Pairs with fewer data points are hidden rather than reported with a misleadingly small denominator.
Not enough back-test data yet
The predictor needs at least three days of past predictions matched against actual prices before we publish a city in the back-test table. Check back tomorrow once the nightly job has had a chance to catch up.
Next step
See where you sit in the cycle
This page shows the wider market. The map adds current station prices, distance, fuel type and directions — so you know whether to fill up now or wait.
Common questions
Predictor accuracy FAQ
Fuel prices move in small increments — a 2-cent miss is a different outcome than a 20-cent miss. MAE measures the actual magnitude of the error rather than a binary right/wrong; it's the standard metric academic and trading-desk back-tests use.
FuelRadar Australia app
Know when to fill up — in your pocket
See the forecast for your fuel and your servos, so you fill up before the next price jump — plus live prices when you need them.
- Today's call: fill up, wait or hold
- A heads-up before the next price jump
- Forecast the bottom of the cycle
- Cheapest servo near you, in seconds
