Estimated energy label of the main building (model, not an energy certificate)

energy_label_estimate13 fieldsNot in the demo

Source: Approach LightGBM model trained on Enova energy certificates and matrikkel building facts
Geography: The address point
Reported stale after: 45 days
Access: Requires a separate agreement (Kartverket Matrikkel)
Contract version: 1
In packages: models

Available with: a separate agreement, switched on per account · Ask about an agreement

What this does not say

Quoted verbatim from the section’s warnings field, exactly as the API returns it. They are not translated: a translation would show something other than what your response actually contains.

Estimate, not an energy certificate (energiattest): under energimerkeforskriften an energy label is a formal document issued for a specific unit, and this is not one. It must never be presented, sold or used as an energy label, for EU-taxonomy reporting or otherwise, without saying it is a statistical estimate. An issued certificate, where one exists, is in energy_certificates and always takes precedence.
A LightGBM classifier with sigmoid-calibrated probabilities, trained on Enova's issued certificates joined to matrikkel facts of the same building: building year and the building code (TEK edition) in force that year, building type, areas, storeys and dwellings, matrikkel energy-source and heating codes, a GRANADA energy well within 50 m and NVE district-heating concession areas, with latitude and elevation as climate proxies (heating degree days are not yet used). estimated_label is the most probable label, confidence its probability, probabilities the full distribution over A-G; the address takes its largest dated building.
The validation block is grouped cross-validation by kommune (each fold scores only kommuner the model did not see): within-one-class accuracy, with exact accuracy, log loss, calibration error and the majority-class baseline beside it. It is null with status not yet computed until the model has been trained on production certificates. Certificates are not a random sample of buildings (sales and new builds dominate), so accuracy on unlabelled buildings can be lower than measured.
Sources: Enova energy certificates (NLOD), Kartverket Matrikkel (by agreement), NGU GRANADA (NLOD), NVE district-heating concessions (NLOD). Model: Approach.

Fields

3 of 13 fields are value fields — the ones that decide whether a record actually says anything. The rest are context: identifiers, codes and timestamps.

Field	Type	Role
method	string \| null	context
method_version	string \| null	context
model_name	string \| null	context
model_version	integer \| null	context
building_id	string \| null	context
estimated_label	string \| null	value
probabilities	object \| null	value
confidence	number \| null	value
expected_label_index	number \| null	context
inputs	object \| null	context
model_trained_at	string \| null	context
validation	object \| null	context
updated_at	timestamp \| null	context

Fetch this section

Send the key in the X-API-Key header. The section can be fetched with the rest of the address, or on its own with pagination.

curl --fail-with-body --silent --show-error \
  -H "X-API-Key: $APPROACH_API_KEY" \
  'https://api.approach.no/api/v1/addresses/285717643?include=energy_label_estimate'

# paginated, one section at a time
curl --fail-with-body --silent --show-error \
  -H "X-API-Key: $APPROACH_API_KEY" \
  'https://api.approach.no/api/v1/addresses/285717643/sections/energy_label_estimate?limit=25'

Why it cannot be tried here

This section is not part of the public demo projection. That is a licensing and privacy decision, not a sign the data is missing: it is served to customers granted the section.