Challenge data

As the 2025 season has been exceptionally extended into 2026, the leaderboard shown reflects exclusively the 2025 results.

Description

Competitive challenge

Industrial

Regression

Time series

10MB to 1GB

Basic level

Dates

Started on Jan. 13, 2026

Challenge context

Challenge goals

Data description

##### Data Structure

•	A dataset x_train.csv with 667,265 rows (i.e., 667,265 stops k, s, d) and 10 columns.
•	A dataset x_test.csv with 20,658 rows (i.e., 20,658 stops k, s, d) and 12 columns.
•	A dataset y_train.csv with 667,265 rows and one column, providing the variable to predict for the training dataset.
•	A dataset y_sample.csv with 20,658 rows and one column with random values between [−10,10]. This is an example submission that you can use as a template to prepare your submission.

##### Data Origin

These data originate from business applications that provide theoretical and observed train arrival and departure times at stations.
The anonymized data are structured as follows: an observation represents a train stop (cle_train) at a given station (cle_gare) on a specific day (date). The variable p0q0 is the difference between the theoretical and actual waiting times. If it is negative, the waiting time is longer than expected; otherwise, it is shorter. The waiting time difference is an integer, as, for simplicity, we only display minutes to our passengers on the screens.

We ask you to predict the difference in waiting time at two stations in real time. We introduce p and q to define the past:

p defines the past values for the previous stops for the same train.
q defines the past values for the previous train at the same station.
The observed waiting time difference is p0q0, which is the difference in waiting time, denoted as Y(k,s,d), for a stop, i.e., a train k at station s on day d.

The columns, i.e., explanatory variables, correspond to 4 contextual variables and 6 past values:

###### Contextual Variables:

-	train (k): train number (unique per day).
-	gare (s): the station identifier.
-	date (d): YYYY-MM-DD, the date the train runs.
-	arret: stop number.

###### Past Variables:

-	p2q0: difference in waiting time for the second previous train k−2 at station s.
-	p3q0: difference in waiting time for the third previous train k−3 at station s.
-	p4q0: difference in waiting time for the fourth previous train k−4 at station s.
-	p0q2: difference in waiting time for the same train k at the second previous station s−2.
-	p0q3: difference in waiting time for the same train k at the third previous station s−3.
-	p0q4: difference in waiting time for the same train k at the fourth previous station s−4.

For more details on the data structure, we invite you to consult our reference article: One-Station-Ahead Forecasting of Dwell Time, Arrival Delay, and Passenger Flows on Trains Equipped with Automatic Passenger Counting (APC) Devices.

Benchmark description

Files

Files are accessible when logged in and registered to the challenge

The challenge provider

Data scientist

PROVIDER WEBSITE

Congratulation for the winners of the challenge

1 Etienne Seckinger
2 Philippe Lonjoux

Challenge Data

Real-time forecast of platform waiting time
by SNCF-Transilien

Description

Dates

Challenge context

Challenge goals

Data description

Benchmark description

Files

The challenge provider

Congratulation for the winners of the challenge

Challenge Data

Real-time forecast of platform waiting time by SNCF-Transilien

Description

Dates

Challenge context

Challenge goals

Data description

Benchmark description

Files

The challenge provider

Congratulation for the winners of the challenge

Real-time forecast of platform waiting time
by SNCF-Transilien