geo_berlin_pv is a large-scale, spatially aggregated geospatial dataset covering all buildings in Berlin, Germany, enriched with socio-economic, building infrastructure, and green technology adoption features. It was constructed to enable GDPR-compliant, privacy-preserving recommender systems for photovoltaic (PV) adoption prediction among private households — without requiring access to entity-level personal data.

The dataset was created and used in the study:

Duran, K. & Monti, A. (2026). Evaluating Recommender System Transferability from Entity-Level Data to Spatially Aggregated Household Data for Green Technology Adoption. Preprint available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7082142

Dataset Details

Key Statistics

Property	Value
Total entries (buildings)	~417,000
Number of features	87
PV adopters (ground truth labels)	~2,680
Geographic scope	Berlin, Germany
Primary data format	Pickle (`.pkl`) / GeoDataFrame
Reference years	2023–2025

Files

File	Description	Size
`0_pv_potential_enriched_labeled.pkl`	Full enriched dataset with all 87 features and PV adoption labels	~393 MB
`1_pv_owner_enriched.pkl`	Subset containing only confirmed PV system owners (ground truth)	~2.47 MB

Dataset Construction

The dataset was assembled by merging multiple openly available geospatial datasets from the Berlin city administration's geospatial data portal. Datasets were merged either on ZIP code or geographic coordinates (GeoPandas polygons), depending on the granularity available for each source.

Source Datasets

Dataset	Reference Year
PV Potential per Building	2025
Number of PV Systems < 30 kW	2025
Number of PV Systems > 30 kW	2025
Population Density	2024
Population Development	2024
Vehicle Information (EV, PHEV, ICE)	2023
Crime Statistics	2023
Building Information (floors, purpose)	—
Building Age (by cohort)	—

Feature Categories

1. Building Data

The base layer of the dataset. Sourced from Berlin's PV potential dataset, it covers every building in the city and includes:

Building purpose (e.g., residential, warehouse, shopping mall)
Roof area and PV potential (possible electricity production, number of modules)
Geographic coordinates
Number of above- and underground floors
Building age cohorts (pre-1900, 1900–1910, …, 2011–2015)
Area structure type (52 categories, e.g., "Freestanding single-family homes with gardens", "Dense inner-city mixed development")

2. Socio-Economic Data

Enriched at the district/ZIP-code level:

Population density and district size
Age distribution of residents (<6, 6–10, 10–18, 18–65, 65–70, 70–75, 75–80, 80+)
Population development (year-over-year change, 2023 → 2024)
Crime statistics: classified as low / medium / high for general crime, domestic/partner crime, and juvenile crime

3. Green Technology Adoption Data

EV adoption: fleet composition (EVs, PHEVs, ICE vehicles), relative shares, new registrations
PV adoption (< 30 kW): aggregate number of modules and installed capacity per area (GDPR-limited resolution)
PV adoption (> 30 kW): precise locations, number of modules, installed capacity, roof orientation, installation date, feed-in type — used as ground truth labels

Intended Uses

Primary Use

This dataset is designed for research on:

Recommender systems for green technology adoption
Spatially aggregated machine learning in GDPR-constrained environments
PV adoption prediction at building or district level
Energy system planning and low-voltage grid expansion

Out-of-Scope Uses

This dataset should not be used to re-identify individual households or persons.
It is not suitable for real-time energy monitoring or grid control.
Predictions should not be used as the sole basis for commercial targeting of individuals.

Data Limitations

Positive-unlabeled (PU) problem: buildings without a recorded PV system are not confirmed non-adopters — they may be in the process of adoption. Recall is therefore the recommended primary evaluation metric.
Temporal mismatch: different sub-datasets have different reference years (2023–2025). Socio-economic conditions may have shifted between reference years.
Building age data: only available up to 2015 cohorts.
Feature completeness: not all available geospatial datasets were exploited; additional features may improve model performance. We invite contributors to add additional information, if available, to enrich the dataset.

Privacy & Compliance

This dataset was specifically designed to be GDPR-compliant:

All data is spatially aggregated at the ZIP-code or district level.
No entity-level personal data (names, addresses, individual household records) is included.
PV systems < 30 kW are only represented in aggregate form, as required by GDPR.

Disclaimer: The data is sourced from public portals and believed to be GDPR-compliant; however, responsibility for data protection conformity lies solely with the original data providers.

Citation

If you use this dataset in your research, please cite the following paper:

@article{duran2026transferability,
  title     = {Evaluating Recommender System Transferability from Entity-Level Data
               to Spatially Aggregated Household Data for Green Technology Adoption},
  author    = {Duran, Kaan and Monti, Antonello},
  journal   = {Energy and AI},
  year      = {2026},
  note      = {Preprint available at SSRN: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7082142},
  doi       = {10.2139/ssrn.7082142}
}

License

This dataset is derived from openly available geospatial data published by the Berlin city administration. The source data is licensed under the Datenlizenz Deutschland – Namensnennung 2.0 (dl-de/by-2.0).

License text: https://www.govdata.de/dl-de/by-2-0

Attribution required: Geodaten Berlin / Senatsverwaltung für Stadtentwicklung, Bauen und Wohnen

This derived dataset is shared under the same terms (dl-de/by-2.0), which is functionally equivalent to CC BY 4.0.

Downloads last month: 21

Total file size:

395 MB