Automated Property Valuation:
Accuracy and Validation
A detailed examination of how the Microburbs valuation system performs across 7,127 Melbourne properties, broken down by price band, property type, and error distribution.
1. Summary
The Microburbs Automated Valuation Model (AVM) estimates residential property prices by combining multiple independent valuation approaches with a broad set of property and market data. This paper presents the validation results from testing on 7,127 residential sales in Melbourne's Outer East region.
Key findings: 62% of valuations fall within 5% of the actual sale price. 87% fall within 10%. The median percentage error is 3.7%. Performance is strongest for properties under $800,000 and weakest for properties over $1.5 million, where data is sparser and individual property characteristics matter more.
2. Overall Accuracy
All results below are from hold-out testing. Each property was valued using only data from other properties. No property ever saw its own sale price during valuation. The testing process was repeated until every property in the dataset had been valued exactly once.
| Metric | Result | What This Means |
|---|---|---|
| Within 5% of sale price | 62% | For a $800,000 property, valuation falls between $760,000 and $840,000 |
| Within 10% of sale price | 87% | For a $800,000 property, valuation falls between $720,000 and $880,000 |
| Within 15% of sale price | 95% | For a $800,000 property, valuation falls between $680,000 and $920,000 |
| Median percentage error | 3.7% | Half of all valuations are off by less than 3.7% |
| Average dollar error | $54,433 | Average difference between valuation and actual sale price |
3. Performance by Price Band
Valuation accuracy varies with property price. Lower-value properties tend to be more homogeneous (similar to their neighbours), which makes them easier to value. Higher-value properties are more unique and harder to compare, so accuracy drops.
| Price Band | Properties | Within 5% | Within 10% | Within 15% |
|---|---|---|---|---|
| Under $600,000 | 1,247 | 68% | 91% | 96% |
| $600,000 to $800,000 | 2,341 | 64% | 89% | 95% |
| $800,000 to $1,100,000 | 2,108 | 60% | 86% | 94% |
| $1,100,000 to $1,500,000 | 934 | 55% | 82% | 93% |
| Over $1,500,000 | 497 | 48% | 76% | 89% |
What drives the difference?
Properties under $600,000 in Melbourne Outer East are typically 3-bedroom homes on standard lots in established suburbs like Bayswater, Boronia, and Ferntree Gully. There are many comparable sales, so the system has strong reference points.
Properties over $1.5 million are often on larger blocks in areas like Warrandyte, Park Orchards, or Mount Dandenong. Fewer comparable sales exist, and individual features (views, land shape, renovation quality) play a larger role. The system still gets within 10% for 76% of these properties, but the 5%-accuracy rate drops to 48%.
4. How This Compares to Typical Valuations
| Aspect | Typical AVM | Microburbs AVM |
|---|---|---|
| Data used | Bedrooms, bathrooms, land size, a few recent nearby sales | Hundreds of property characteristics, comparable sales at multiple distances, listing history, area-level market activity |
| How it handles missing information | Fills in an average value | Treats missing data as a signal (a property with no nearby sales is different from one with many) |
| Number of valuation approaches combined | One | Multiple independent approaches blended together |
| Error checking | Manual review of results | Automated checks flag suspicious data before it reaches the valuation system |
| Accuracy by our testing | Rarely published | 62% within 5%, 87% within 10%, tested on 7,127 properties |
The core difference is data breadth. Most automated valuations rely on a small number of features (typically under 20). Our system draws on a much larger set of property and market characteristics to capture what makes each property distinct.
5. Head-to-Head Comparison with Domain
We tested our valuations directly against Domain across 7,037 Melbourne properties where both systems provided an estimate. Both were compared against the same recorded sale prices from the land titles office.
| Metric | Microburbs | Domain |
|---|---|---|
| Within 5% of sale price | 60.1% | 49.3% |
| Within 10% of sale price | 85.2% | 85.7% |
| Median dollar error | $37,626 | $49,799 |
| Closer on individual properties | 60.6% | 39.4% |
At the tightest accuracy band (within 5%), the Microburbs AVM leads by 10.8 percentage points. Property-by-property, our estimate was closer to the sale price on 60.6% of cases. The median dollar error is $12,173 smaller per property.
Domain edges ahead at the within-10% mark by 0.5 percentage points and is more accurate on properties above $1.2 million. In the $600,000 to $1 million range, where the majority of investors operate, the Microburbs AVM was closer on 68-74% of individual properties.
Timing of estimates
Every MicroVal prediction has a recorded date in our database. Each was generated before the property sold. Domain's estimates were collected from their website on 16 January 2026. By that date, almost every property in the comparison had already sold (most sold between January 2023 and December 2025).
We cannot confirm whether Domain updated their estimates after each sale was recorded. This timing difference should be considered when reading the results. A valuation made before a sale is a genuine prediction. A valuation that could have been updated after a sale may reflect the outcome rather than predict it.
6. How We Validated These Results
We used a standard hold-out testing approach. The 7,127 properties were split into five groups. For each group, we valued those properties using only data from the other four groups. This process was repeated until every property had been valued once, without ever seeing its own sale price.
This is the same validation approach used in academic research on property valuation. It prevents the system from memorising answers and gives an honest measure of accuracy on unseen properties.
6.1 Data Integrity
Every data source goes through automated integrity checks before it reaches the valuation system. These checks look for data that is too closely correlated with the sale price (which would mean the system is accidentally seeing the answer), values that cluster suspiciously at recent dates, or any single data point with outsized influence on the result.
During development, one data source initially inflated accuracy to 90%. Our automated checks flagged it within hours: the field contained the actual sale price for 14% of properties, labelled under a different name. We removed it and reported the honest number.
Why this matters
Many valuation providers report accuracy numbers without disclosing how they were measured. Were the same properties used to build and test the system? Were outliers removed? Were results cherry-picked from the best-performing region? We report results on all 7,127 properties with strict hold-out testing. No exclusions, no cherry-picking.
7. Where the System is Less Accurate
No valuation system is perfect. Ours struggles most in these situations:
- High-value properties ($1.5M+): Fewer comparable sales, more unique features. Within-10% drops from 91% to 76%.
- Recently subdivided lots: The system has limited data on newly created addresses and may over-value based on the original property.
- Properties with unusual features: A property with a tennis court, granny flat, or commercial zoning may not have enough similar sales for an accurate comparison.
- Rapidly changing markets: In areas where prices are moving quickly (up or down), the system may lag behind the most recent trends.
For these cases, the valuation should be treated as a starting point, not a definitive answer. Combining it with a local agent appraisal gives the best result.
8. Conclusion
The Microburbs AVM achieves 62% within-5% and 87% within-10% accuracy on 7,127 Melbourne Outer East properties. Performance is strongest under $800,000 and weakest above $1.5 million.
The system draws on a broad set of property and market data to capture what makes each property distinct. Strict hold-out testing and automated integrity checks ensure reported accuracy reflects real-world performance.
Future work will extend testing to additional regions across Australia and improve accuracy for high-value and data-sparse properties.
Want Property Valuations You Can Trust?
Microburbs property valuations are backed by rigorous testing across thousands of actual sales.