{ "fetched_at": "2026-10-19T06:39:43Z", "source_url": "https://www.kaggle.com/competitions/gait-data-processing/data", "pages": [ { "name": "data-description", "content": "The data for this competition consist of EEG and inertial signals, which were recorded with sampling rate at 512 Hz and 20 Hz, respectively, while patients with complete spinal cord injury (SCI) walked under the Lokomat's assistance. The inertial signals were resampled at 512 Hz, and together to the EEG data are provided in CSV files, which later were split into a training set and a testing set. \n\n## Training Data\nThe file \"train_set.csv\" contains the EEG recordings that were collected in several sessions from two SCI individuals. The file is composed of columns, indicating the following information:\n1) ID: Each row contains a unique identifier;\n2) Subject: The SCIX indicates the subject;\n3) Session: The number of each session takes values from 0001 to 00XX;\n4) Time:512Hz: The time sequence.\n5) 16 EEG locations, according to the international 10-20 system: FP1, FP2, F3, F4, FC1, FC2, C4, C3, Cz, CP2, CP1, P4, Pz, P3, O2, and O1. \n\n## Labels of training\n1) ID: Each row contains the same unique identifier that corresponds to the training set;\n2) Angle: The inertial signal recorded simultaneously with the EEG data.\n\n## Testing Data\nThe file \"test_set.npy\" contains EEG epochs of 1 s (16 channels x 512 samples) extracted each 0.1 s from unseen EEG recording of two SCI individuals. This file has the following structure:\nTo submit the results, the competitors of this challenge are required to predict the inertial signal values from the testing set, and follow the instruction below.\n\n## Example for submission\nThe provided file \"example_submission.csv\" contains the required format to submit the predictions from the testing set. Notice that this file contains two columns with ID and prediction, where ID contains the same unique identifier corresponding to the testing set, and prediction contains the predicted inertial signal values.\n\nIt is worth noting that the training and testing sets were randomly separated from the full dataset, taking into account the session number. The competitors are allowed to use external data sources for their models, as long as they are publicly available and properly cited.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Description", "content": "# Goal of the Competition:\nDevelop a machine learning model to decoding the gait phase by using electroencephalogram (EEG) data collected from a person with spinal cord injury using the Lokomat system. As a reference, the signal from a inertial unit motion (IMU) sensor is used.\n\n# Motivation\nWearable devices, such as robotic exoskeletons, have been used to assist spinal cord injury (SCI) individuals in regaining some of their lost mobility. These devices can be further enhanced by incorporating the individual's intended movement via motor imagery (MI) detected through EEG. To promote the development of more accurate and efficient machine learning models for decoding joint values from EEG data, we propose here a competition that utilizes data collected from SCI individuals who walked on a treadmill with the Lokomat's assistance.\n\n# Methodology:\nTo record EEG signals, 16 wet electrodes were placed on the scalp according to the international 10-20 system (FP1, FP2, F3, F4, FC1, FC2, C4, C3, Cz, CP2, CP1, P4, Pz, P3, O2, and O1). The Fpz and Fz electrodes were used as ground and reference, respectively. The electrode's impedance was kept below 10 kΩ. Signals were acquired with the V-Amp 16 system at 512 Hz sampling rate. An in-house inertial sensor, termed JAMA, was placed between the calf and ankle braces to acquire the gait phases by using sampling rate at 20 Hz. Both raw EEG and IMU signal were synchronized for analysis. The participant was positioned on Lokomat, and received 100% assistance walking at 1 km/h with body weight support. \n\nThe Lokomat robotic system is designed to assist and control lower-limb movements during gait rehabilitation. It is equipped with a sophisticated software system that allows the therapist to program various walking patterns and levels of assistance, which can be tailored to the individual needs of each patient. The Lokomat also provides body weight support and an adjustable speed control.\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F10622352%2F4d7955623ff1e61153cbce9403e9a029%2FLogos%20assinatura.png?generation=1681933899167807&alt=media)\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Evaluation", "content": "This competition aims to develop a machine learning model to predict inertial signal values from EEG data collected from a person with spinal cord injury using a Lokomat exoskeleton. The predictions will be evaluated based on their **root mean squared error (RMSE)** between the predicted and actual values.\n\n## Submission Format\nParticipants are required to submit their predictions in a CSV file with the following format:\n\nThe file should contain a header row with two columns: \"id\" and \"prediction\".\nThe \"id\" column should contain unique identifiers for each test sample.\nThe \"prediction\" column should contain the predicted inertial signal values for each test sample.\n\n## Evaluation\nSubmissions will be evaluated using the mean squared error (MSE) between the predicted and actual inertial signal values. The formula for the MSE is as follows:\n\n$$RMSE = \\sqrt{1/N * sum((y_{pred} - y_{true})^2)}$$\n\nwhere N is the total number of test samples, $$y_{pred}$$ is the predicted value, and $$y_{true}$$ is the actual value.\n\nThe score for each submission will be calculated based on the RMSE, and the submission with the lowest MSE will be ranked at the top of the leaderboard. The final ranking will be based on the scores of the top five submissions.\n\nNote: In the event of a tie, the submission with the lower run time will be ranked higher.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "rules", "content": "Each team must have at least one and the maximum of five participants;\nOnly kaggle submissions with its code will be considered for publication in the journal.\nWinning teams of each challenge will be invited to present the results at the IX Neuroengineering Symposium.\n", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Timeline", "content": "- **June 1, 2023** - Start Date.\n\n- **September 1 , 2023** - Entry Deadline. You must accept the competition rules before this date in order to compete.\n\n- **September 1 , 2023** - Final Submission Deadline.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false }, { "name": "Paper Publishing Opportunity", "content": "The top-performing **code submissions** will have the opportunity to publish a methodological paper.\n\nSelected teams will be contacted and asked to submit a manuscript within 30 days of the competition's completion. The manuscript will be evaluated by the Scientific Committee and, if accepted, sent for publication.\n\nPlease note that although the competition will be accepting submissions with and without code, **only submissions with accompanying code will be considered for publication**. This is to ensure that the methods presented in the manuscript are transparent and reproducible, as well as to encourage open science practices in the field of neural engineering. We encourage all participants to carefully review the competition rules and guidelines and to submit their best work for consideration.", "isPublished": false, "hasIsPublished": false, "postTitle": "", "hasPostTitle": false, "mimeType": "", "hasMimeType": false } ] }