Skip to content

bbdc_2025_description_en.md

Latest commit

 

History

History
85 lines (51 loc) · 5.96 KB

bbdc_2025_description_en.md

File metadata and controls

85 lines (51 loc) · 5.96 KB
Raw

Bremen Big Data Challenge 2026

Task Description

The Bremen Big Data Challenge (BBDC) 2026 focuses on the detection of age-related diseases based on multiple physiological signals in everyday situations. The physiological data were recorded while study participants performed everyday tasks in a virtual reality environment simulating various diseases.

Participants in the competition receive data from "healthy" individuals as well as data from six diseases with varying severity levels. The dataset includes electrodermal activity (EDA), electroencephalography (EEG), and electrocardiogram (ECG) data. EDA data captures changes in skin conductance, EEG data measures brain activity, and ECG data records the electrical activity of the heart. Together, these signals provide insights into diverse physiological states of the human body. The task of BBDC 2026 is to identify these states.

For the Student Track, pre-extracted features are additionally provided as a starting point. However, all participants are encouraged to generate additional features or to process the physiological data in its raw form. This is because the provided features are not necessarily informative enough to solve the challenge optimally.

Furthermore, participants in the Student Track are provided with the severity levels of the diseases for the training data. These are not available for the test data, nor are Student Track participants expected to predict them. They serve only as additional information for analyzing the data. Disease severity ranges from 0 (no disease) to 4 (severe disease).

Participants of BBDC 2026 are required to read and comply with the data license agreement (see Data_Licence_Agreement.txt).

There is one folder for training (train) and one for testing (test).

Training

The training folder contains seven subfolders. Six of them correspond to the different diseases, and the seventh contains data from "healthy" individuals. Each of these subfolders contains four files:

  1. data.csv
  2. features.csv
  3. labels.csv
  4. stages.csv

A version of each file exists for each class to be predicted. We also provide example code that combines the data for all classes into a single table (see combine_files.py).

The structure of the files is described below.

  1. data.csv: Contains all raw EEG, ECG, and EDA training data for the respective class. All signals are sampled at 250 Hz and divided into 10-second segments. Each segment has a unique identification number.
Column Description
ID Unique segment ID.
EEG_CZ, EEG_FP2, EEG_F3, EEG_FZ, EEG_F4, EEG_T7, EEG_C3, EEG_FP1, EEG_C4, EEG_T8, EEG_P3, EEG_PZ, EEG_P4, EEG_PO7, EEG_PO8, EEG_OZ The 16 EEG channels. The naming and placement follows the international 10-20 system.
ECG The ECG data.
EDA The EDA data.

  1. features.csv: Some pre-extracted features. These can serve as a starting point for model development. Each row represents the features for one segment. For the EEG data, we provide different frequency bands (delta, theta, alpha, beta, and gamma). For the ECG and EDA data, we compute the mean, standard deviation (std), minimum (min) and maximum (max) values including ranges (ptp), the effective signal energy (rms), and the maximum temporal slope (max_slope) of each signal.

  2. labels.csv: This file indicates which disease is present in the respective segment. These values are the target variables for model training:

Column Description
ID Unique segment ID.
impairment_type An integer in the interval [0, 6], where 0 represents the "healthy" class and 1–6 represent the disease types.
  1. stages.csv: The severity levels of the diseases for the training data of the respective class. They serve as additional information for data analysis.
Column Description
ID Unique segment ID.
impairment_stage An integer in the interval [0, 4], where 0 represents the "healthy" class and 1–4 indicate the disease severity.

Testing

The test folder contains three files:

  1. data.csv
  2. features.csv
  3. student_skeleton.csv

The details of these files are described below.

  1. data.csv: Contains the raw EEG, ECG, and EDA test data for all classes combined into a single table. All signals are sampled at 250 Hz and divided into 10-second segments. Each segment has a unique identification number. The fields correspond to those described in (1).

  2. features.csv: Some pre-extracted features of the test data. Each row represents the features for one segment. The fields correspond to those described in (2).

  3. student_skeleton.csv: This file must be filled with the predicted diseases. The fields correspond to those described in (3). The number of rows and the order of the IDs must not be changed.

Submission

Once the file student_skeleton.csv has been completed, it can be uploaded to the BBDC 2026 submission portal (https://bbdc.csl.uni-bremen.de/submission/). The score will be calculated and displayed automatically, and the leaderboard will be updated.

Evaluation

The final score is calculated based on the Macro-F1 score for disease prediction. The minimum score is 0.0 (0%), and the maximum score is 1.0 (100%). A higher score is better.