Dataset and Code for 'Feature Extraction-based Clustering Selection Methodology to Identify Representative Buildings for Scalable Energy Simulations'

This repository contains the anonymized dataset and reproducible code accompanying the paper:

“Feature extraction-based clustering selection methodology to identify representative buildings for scalable energy simulations.”

The dataset includes:

• epc_metadata.csv: EPC-based building attributes (energy performance value, weighted U-value, air leakage, heat recovery)

• dh_merged_kwh_per_m2.csv: hourly district heating consumption, normalized per m²

• outdoor_temp_2023.csv: hourly outdoor temperature data (2023)

• features_epc_only_standardized.csv: standardized feature table used for clustering

• data_dictionary.xlsx: description of all variables

The code provides a minimal pipeline to reproduce the main results of the paper:

1. Load and preprocess data

2. Extract EPC-based features

3. Run clustering (K-Medoids, Agglomerative, GMM)

4. Compute validation metrics and agreement indices

Expected outputs (e.g., validation_metrics.csv, agreement_metrics.csv) reproduce the main tables in the manuscript.

Anonymization: all identifiers (school names, locations, addresses) are removed. Only numerical features are included.

Licenses:

• Code: MIT License

• Data: Creative Commons Attribution 4.0 International (CC BY 4.0)

Related resources:
The GitHub repository with the same reproducible pipeline is available at:
https://github.com/hatefh/zenodo_code_bundle_epc_clustering_paper