Load-Climate-Open: an open hourly dataset and reproducible pipeline for day-ahead electricity-load forecasting with climate and renewable-generation features

A reproducible hourly dataset for day-ahead electricity-load forecasting, enriched with climate (air temperature, heating/cooling degree-hours), calendar and renewable-generation features. It is the fully-open, redistributable counterpart of the author's electricity-consumption-forecasting work (KMOSS-2025 thesis); both upstream sources are CC-BY 4.0 (Open Power System Data / ENTSO-E, and Open-Meteo ERA5), so the entire derived dataset is deposited openly. Two aligned versions: base_v1 (a pure calendar model, 5 features) and improved_v2 (autoregressive load lags + climate + renewable-generation lags, 13 features). A regression family (Linear, KNN, Decision Tree, GBM, MLP) is trained with a strictly chronological split; every model improves, the best (GBM) reaching MAPE 2.77% / R2 0.939 (vs 4.25% / 0.872 for the calendar baseline) — in the ~3% range of the recurrent/transformer models in the KMOSS-2025 study, on a fully-open dataset. Causal features only (no leakage); strictly chronological split. The data are the open German (DE) zone series (OPSD/ENTSO-E + Open-Meteo), 2015-2019, hourly, used as a fully-redistributable stand-in for the non-redistributable Ukrainian series of the thesis. Supplements the author's KMOSS-2025 thesis (Shapovalova S.I. & Titov V.M., 2025, "Forecasting electricity consumption in Ukraine based on neural-network approaches").