Vietnam Robusta Coffee Price Forecasting Dataset with RON95 Fuel and Weather Features, 2021-2026
Description
This dataset contains raw and processed data used for coffee price forecasting in Vietnam. It includes Robusta coffee prices collected from public agricultural market sources, RON95-III Region 1 fuel prices collected from Petrolimex announcements, and historical weather variables retrieved from Open-Meteo. The dataset also provides cleaned daily files, merged modeling data, and time-based train/validation/test splits for reproducible forecasting experiments.
Files
Steps to reproduce
This dataset supports reproducible experiments for forecasting Robusta coffee prices in Vietnam using public data sources. It includes raw, cleaned, merged, and modeling-ready files used in the associated thesis experiments. The workflow contains five main steps. First, raw coffee price data were collected from Vietnamese agricultural market information sources. Fuel price data were collected from Petrolimex public fuel price adjustment announcements, focusing on RON95-III Region 1. Weather data were retrieved from Open-Meteo using province-level geocoding information for Vietnamese coffee-producing areas. Second, the raw files were cleaned and standardized. Dates from different sources were converted into a consistent daily format. Coffee prices were standardized by date and market/province. Fuel prices were converted into a daily time series using official adjustment dates and forward filling between updates. Weather records were aligned by date and province. Third, the cleaned coffee, fuel, and weather files were merged to create modeling-ready datasets. These files allow different feature combinations, including coffee price only, coffee price with fuel price, coffee price with weather variables, and full combinations of coffee price, RON95 fuel price, precipitation, and temperature. Fourth, the data were split according to chronological order instead of random sampling to avoid future-data leakage. The dataset provides predefined train, validation, and test/holdout splits for both a single-province setting and an extended multi-market setting. Fifth, the provided result tables summarize model evaluation on the holdout set using RMSE, MAE, MAPE, and MASE. Users can reproduce the experiments by starting from the raw file, inspecting the cleaned files, using the merged data as the modeling base, applying the predefined splits, and comparing outputs with the result tables. All data sources are public. The included files should be treated as a fixed snapshot for reproducibility, because recollecting data from the original websites may produce slightly different records due to later updates or website changes.