FI-TW: An open train–weather dataset for railway delay analysis in Finland

Train delays result from complex interactions between operational, technical, and environmental factors. Weather strongly affects railway reliability in Nordic regions, yet publicly available datasets rarely integrate meteorological information with operational train records, which limits research on weather-driven delay.

We constructed the Finland Integrated Train-Weather (FI-TW) dataset by combining operational records from the Finland Digitraffic Railway Traffic Service with observations from 209 Finnish Meteorological Institute stations, covering January 2018 to December 2024. Train events and weather measurements were aligned in space and time using the Haversine distance, with a radial fallback strategy that recovers missing parameters from alternative nearby stations. Processing included cyclical encoding of temporal features, robust scaling of weather data to limit the effect of sensor outliers, duplicate removal, and the derivation of weather-scenario indicators together with multi-scale rolling-window aggregations.

The dataset contains 138 features spanning operational variables and meteorological measurements, and approximately 38.5 million observations from Finland’s 5,915-kilometer rail network. Exploratory analysis shows a clear seasonal structure, with winter delay rates exceeding 25% compared with below 20% in summer, and geographic clustering of high-delay corridors in central and northern Finland. A baseline experiment using extreme gradient boosting (XGBoost) regression reached a mean absolute error of 2.73 minutes for station-specific delay prediction.

FI-TW is, to the best of our knowledge, the first publicly available dataset that couples Finnish railway operations with synchronized meteorological observations over a seven-year span. It offers multiple target variable formulations and supports applications such as delay prediction, weather impact assessment, seasonal reliability analysis, and infrastructure vulnerability mapping, providing a reusable resource for machine learning research on railway operations in severe northern climates.

Trains in Finland often run late, and the weather is one of the reasons. Snow, ice, and very low temperatures can damage equipment, slow trains down, and disrupt signalling, while in autumn fallen leaves make the rails slippery. Researchers who want to measure how much the weather really matters need data describing both the trains and the weather at the same place and the same moment. Until now that combination has been hard to find, because public railway datasets usually describe only train movements and weather datasets usually describe only the atmosphere.

This work presents FI-TW, a free and open dataset that joins the two. It brings together seven years of Finnish train records, from January 2018 to December 2024, with measurements from 209 weather stations across the country. For every point where a train passes or stops, the dataset reports how early or late it was and what the weather was doing nearby at that moment. Where a nearby weather station was missing a measurement, the value was taken from the closest station that recorded it.

The result is a single collection of about 38.5 million records, each described by 138 pieces of information. In the data, more than one train event in four is delayed during the winter months, compared with fewer than one in five during summer, and the most affected routes lie in central and northern Finland. As a demonstration, a standard machine learning model trained on the data predicted station arrival delays with an average error of under three minutes.

The dataset is free for anyone to download and reuse. It can support work on predicting delays, measuring how much specific weather conditions cost the railway, planning timetables that account for the season, and identifying the parts of the network most vulnerable to extreme weather.