One sentence in, a price estimate and its evidence out

There are no dropdowns to work through. Type the address, the size and the property type the way you would say it out loud:

Abbreviations are expected — JLN for Jalan, TMN for Taman, and the shorthand people actually use for property types. Malay and English both work.

Before estimating anything, it shows what it understood, so a misreading is visible rather than hidden inside the result:

Two of those were not in what you typed, and the tool says so rather than quietly filling them in:

• Floor area — you gave one size, 143 sq m. The tool treated it as the land area, because 143 sq m is exactly the median land size for terraced houses in this neighbourhood, and took the neighbourhood's median floor area of 130 sq m for the built-up figure.
• Tenure — not stated, so it assumed freehold, which is what almost all nearby sales are.

It also reports what it matched against: Taman Johor Jaya, 962 recorded sales, and 59 sales on Jalan Bakawali itself. If your scheme had only a handful of records you would see that too, and know to treat the estimate loosely.

This is the part that matters most, and it follows the way a valuer actually works: the nearest evidence first, widening only when there is not enough of it.

1. Same road first. If the exact road has two or more recorded sales of the same property type, only those are used. Nothing else comes close as evidence.
2. Then the road family. Jalan Bakawali 8, Jalan Bakawali 10, Jalan Bakawali 18 and so on are treated as one group — they are typically the same developer, the same era, the same house design.
3. Then the scheme, then the mukim. Each step out is a step down in reliability, and the tool tells you which one it used.

For this property it stopped at step one. Three sales on Jalan Bakawali 8 itself:

These are real registered transactions from the NAPIC dataset, shown to you in full. You can see their dates, their sizes and their prices, and judge for yourself whether they are fair comparisons for the property you are asking about. Notice the middle row: same road, same land area, same floor area as the subject. That is about as close as open transaction data gets.

Separately from the comparables, a gradient-boosted regression model (LightGBM) trained on 92,000 Johor transactions predicts a price from the property's type, size, scheme, tenure, location coordinates and the transaction date. For this property:

The range is not decoration. It comes from quantile models trained alongside the main one, and it is wide because property prices genuinely are: two houses on the same road with the same floor plan routinely differ by RM100,000 on condition and renovation alone, and none of that is in the data.

Sanity-check it against the evidence above. The three Jalan Bakawali 8 sales ran RM520,000 to RM590,000 during 2023. An estimate of RM576,000 for 2026 sits just above that band — consistent with a market that has risen since, and with the middle comparable being the closest match in size.

How accurate is it?

Tested honestly — trained only on transactions up to 2024, then asked to price 8,665 sales from 2025 that it had never seen — the median error was 12.2%. Half of all estimates land within 12.2% of the actual price; the other half are further out, sometimes considerably. On a RM576,000 estimate that is roughly RM70,000 either way at the median.

That is useful for orientation and useless for a decision that turns on the exact figure. It is why this site says "price estimate" and never "valuation".

The estimate comes with the subject property and each comparable plotted, at the best precision the data allows — and labelled with which precision that is:

• Road level when the road name geocodes to a real street. Most landed housing lands here.
• Building level for high-rise, when the block name is recognised.
• Scheme approximate when only the taman can be located — drawn deliberately as an area rather than a pin, so it does not imply precision the data cannot support.

The separate Price Explorer works the other way round: click anywhere in Johor and it summarises every recorded sale within a radius you choose, from 1 to 10 km. Useful when you are asking "what does this area cost?" rather than "what is this house worth?".

What it deliberately does not do

• It does not value your renovation. The data records what a property is and what it sold for, never its condition, its floor position or its view.
• It does not use asking prices. Everything here is registered transacted prices — what buyers actually paid, not what sellers hoped for.
• It does not pretend to precision it lacks. Small samples are labelled, assumptions are shown, the range is published alongside the estimate.
• It is not a valuation. A formal valuation involves a registered valuer inspecting the property. This is a statistical estimate from public data, and the two are not substitutes.

Where the data comes from

All 92,000 transactions come from NAPIC (JPPH Malaysia) open transaction data, covering Johor from 2021 to the latest published quarter. Coordinates come from OpenStreetMap via Nominatim, cached so the same place is not looked up twice. The dataset is refreshed each time NAPIC publishes a new quarter, and the model is retrained with it — the quarterly reports track what changes.

Price estimator · Market reports · Buyer's guide · How to read price per square foot · About & contact