Accuracy

How XAI Land Measures and Validates Its AVMs


1. What Does “Accuracy” Mean?

An Automated Valuation Model (AVM) estimates a property’s value based on real market data.

Accuracy means:

How close our estimated value is to the actual transaction price recorded by Ministry of Land, Infrastructure and Transport (MOLIT).

We measure this using internationally recognized statistical standards — not marketing claims.

What is AVM Accuracy?
What is AVM Accuracy? AVM Estimated Value ₩950,000,000 Actual Transaction Price (MOLIT) ₩1,000,000,000 |Estimated – Actual| ÷ Actual = Error % |₩950M – ₩1,000M| ÷ ₩1,000M = 5% Absolute Percentage Error 5%

2. How We Build Our AVMs

Our valuation models are built in four structured steps:

4-Step AVM Development Process
AVM Development Process STEP 1 Data Collection Transaction records, property attributes, market data STEP 2 Data Engineering Cleaning, feature extraction, quality assurance STEP 3 AI Model Training Algorithm selection, hyperparameter optimization STEP 4 Out-of-Sample Validation Independent testing, accuracy measurement

Step 1 – Comprehensive Data Collection

We collect nationwide data including:

  • Official MOLIT transaction prices
  • Apartments, Villas, Officetels
  • Detached & Multi-family homes
  • Small commercial buildings (꼬마빌딩)
  • Land parcels
  • Infrastructure, transit, schools, hospitals
  • GIS & environmental data

Step 2 – Data Engineering & Quality Control

Before modeling, we:

  • Geocode properties
  • Remove outliers
  • Merge datasets
  • Eliminate duplicates
  • Handle missing values
  • Construct structured modeling features

Only clean, structured data enters our models.


Step 3 – AI Model Training

We train machine learning models using:

  • Sales comparison methodology
  • Market-based valuation logic
  • Region-specific calibration

Each property type is modeled independently.


Step 4 – Out-of-Sample Validation

We test our models against:

Real MOLIT transaction data the model has never seen before.

This ensures objective performance measurement.


3. How We Measure Accuracy

We use transparent statistical metrics.

1️⃣ MdAPE (Median Absolute Percentage Error)

Measures the median percentage difference between our estimate and actual transaction price.

Lower = More Accurate.

Why median?
It reduces distortion from extreme outliers.

MdAPE Explanation
Median Absolute Percentage Error (MdAPE) Lower is Better 5% Property 1 8% Property 2 12% Property 3 MEDIAN 20% Property 4 40% Property 5 The median represents the middle value when errors are sorted. It is less sensitive to extreme outliers than the mean.

View our latest accuracy results:


2️⃣ PPE @% (Percentage Predicted Error) (±5%, ±10%, ±20%)

PPE@10% shows the percentage of valuations that fall within ±10% of the actual transaction price.

Higher PPE indicates stronger reliability.

We also evaluate additional thresholds (±5%, ±20%) for internal monitoring.

Accuracy Rate Bands
Accuracy Rate 0% 25% 50% 75% 100% XX% Within ±5% XX% Within ±10% XX% Within ±20% % of Valuations

3️⃣ Hit Rate & Record Count

  • Hit Rate: % of properties successfully valued
  • Record Count: Size of our accessible valuation database

High coverage + high accuracy = reliable infrastructure.


4. Why Accuracy Matters

AVM accuracy is not just a technical metric.

It directly impacts:

  • Loan-to-Value (LTV) risk
  • Collateral monitoring
  • Provisioning sensitivity
  • Capital allocation decisions
  • Systemic financial stability

Transparent accuracy measurement reduces overvaluation risk and improves responsible lending.

Why Accuracy Matters
Why Accuracy Matters The Financial Risk Chain STEP 1 Property Value STEP 2 Loan-to-Value (LTV) STEP 3 Credit Risk STEP 4 Capital Allocation Accurate property valuation is the foundation of sound lending decisions and financial stability

5. Continuous Improvement

Our models are:

  • Regularly recalibrated
  • Continuously retrained
  • Performance-monitored
  • Documented for audit review

Accuracy is not a static number.
It is a monitored and managed process.

Our AVM accuracy is directly linked to the strength of our data pipeline and infrastructure.

We continuously improve our data ingestion, cleaning, feature engineering, geocoding, and model retraining workflows to ensure reliability and scalability.

For readers interested in our technical data architecture and pipeline evolution:

🔗 Learn more about our enhanced data pipeline


6. Research & Policy Contribution

XAI Land actively contributes to the development of transparent AVM standards in South Korea.

Our CEO co-authored a peer-reviewed paper published in the Journal of Real Estate Analysis (2024):

“자동가치산정의 정확성 분석 및 투명성 관리 방안”
(An Analysis of the Accuracy of Automated Valuation Model and Transparency Management Plan)

The study:

  • Analyzed nearly 20,000 property transactions
  • Evaluated AVM performance using MdAPE and PPE@10%
  • Demonstrated regional variation in accuracy
  • Highlighted the importance of region-specific standards
  • Proposed independent third-party validation frameworks
  • Discussed AVM’s role in reducing overvaluation risk and preventing financial instability

This research reflects our commitment to measurable standards and transparent validation.

🔗 View Published Paper (JREA, 2024)


7. Institutional Reporting

Detailed regional accuracy reports are available for:

  • Banks
  • Financial regulators
  • Public institutions
  • Institutional investors

Contact us to request a full technical accuracy report.