Participants will work with hourly time series data that includes in-store transactions from two retailers located in Würzburg’s historic city center. The prediction task is to forecast the **hourly customer counts** for **Company A** and **hourly sales across three product categories** for **Company B** over a two-week test period.

To support more accurate forecasting, participants are provided with pedestrian foot traffic data from three nearby streets: Schoenbornstrasse (green), Kaiserstrasse (red), and Spiegelstrasse (blue). These streets are equipped with laser-based counters that continuously measure pedestrian movement and were featured in the previous [Foot Traffic Würzburg](https://www.kaggle.com/competitions/foot-traffic-wue/) competition.

While the previous competition focused on developing accurate models to forecast the hourly number of pedestrians in Würzburg’s city center, this challenge takes the next step: foot traffic now serves as auxiliary information to support the prediction of customer counts and sales. This reflects the broader vision from the outset of this project — to ultimately use pedestrian data to improve forecasting for retailers and thereby support more effective operational and planning decisions.

To help participants, it is specified which of the three foot traffic counters is closest to each company. However, for privacy and legal reasons, we are not permitted to publish detailed information about the businesses involved. Accordingly, the data has been anonymized to protect the identity of both companies.

![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F22956516%2Fa561867dbe1950c848a32932bf359aba%2FStraenmessungen_Wurzburg_resized.png?generation=1746522933173881&alt=media)

The foot traffic data includes hourly pedestrian counts split by direction:

* `n_pedestrians` – total pedestrian count
* `n_pedestrians_towards` – pedestrians walking **towards** the city center
* `n_pedestrians_away` – pedestrians walking **away** from the city center

**Important:** For the **test period**, the actual foot traffic data is **not provided**—neither for the public leaderboard nor the private leaderboard test sets. Instead, only forecasts of the foot traffic can be used during this period to simulate the real-world setting, where the true foot traffic of future time points is also not available for planning purposes. Instead of requiring participants to build their own foot traffic forecasting models, we provide the **predictions generated by the best-performing model** from the previous [Foot Traffic Würzburg](https://www.kaggle.com/competitions/foot-traffic-wue/) competition. These forecasts can be used directly as auxiliary inputs to support sales and customer count predictions. Participants may also choose to generate their own foot traffic forecasts if they believe they can improve upon the provided baseline.

In addition, participants have access to **contextual information** such as:

* **Weather data**, including temperature, rainfall, and other conditions, is provided together with the foot traffic data.
  → To simplify the forecasting task, **weather data is provided even for the test period**, removing the need to independently predict future weather.
* A curated list of **local events** (e.g., festivals, markets, city-wide celebrations)
* **Public holidays** and **school holidays** in Bavaria
* Data on the **lecture periods** of the University of Wuerzburg.

These features are expected to play a key role in explaining short-term deviations in customer activity.

Participants may incorporate additional **external data**, as long as it was publicly available **before the start of the test period**. In particular, **economic indicators**—such as inflation data—may offer valuable signals to improve forecasting performance.

The locations of the foot traffic counters are also included in the data. For interactive exploration, see the [Open Data platform of the City of Würzburg](https://opendata.wuerzburg.de/explore/dataset/pedestrians_counter_geojson/map/?location=16,49.79659,9.93497&basemap=jawg.streets), where the measurement zones are visualized.