Participation | Evaluation Guidelines | Paper Submission | Dataset Information | Tutorial Notebook | FAQ

Tremor Challenge

Tremors are one of the most common and noticeable motor symptoms of Parkinson's disease, categorized as unusual activities due to their involuntary and rhythmic movements. These tremors often manifest as resting tremors, occurring when the muscles are relaxed and disappear during voluntary movement or sleep. Tremors typically affect the hands, fingers, legs, or jaw, significantly impacting the patient's ability to perform everyday tasks, such as writing, eating, or holding objects.
Parkinson's disease (PD) is a slowly progressive nervous system condition caused by the loss of dopamine-producing brain cells. It primarily affects the patient's motor abilities, but it also has an impact on non-motor functions over time. Patients' symptoms include tremors, muscle stiffness, and difficulty walking and balancing. Then it disrupts the patients' sleep, speech, and mental functions, affecting their quality of life (QoL).
In the context of understanding Parkinson’s disease through sensor data, tremors serve as a critical indicator of disease progression and response to treatments. By analyzing various modalities, such as motion sensor data, or video recordings, researchers can gain deeper insights into the frequency, amplitude, and patterns of tremors. Such insights are essential for distinguishing between normal and unusual activities, enabling more precise diagnoses, treatment personalization, and monitoring of therapeutic outcomes.

Challenge Goal

To recognize some normal and unusual/tremor activities from sensor data, towards progressing our research activities globally – related to wearing-off of Parkinson’s Disease (AD), Alzheimer’s Disease (AD), dementia which is extremely important for elderly life globally.

Let’s DO the BEST!

Item Deadline
Team Registration (Registration Link (Select: "Tremor Challenge")) by 10 May
Results Submission by 22 May
Paper* (method and analysis) Submission by 24 May
PD Challenge Workshop 28 May
*Accepted papers will be published in the IJABC Journal. Format: Overleaf Template.

Participation

The challenge is open to both students and working individuals.

A. Registration Requirements

Mandatory Registration: All participants, including solo challenger, must complete and submit the Team Registration Form prior to joining the competition:

CMT: Registration Link (Select: "Tremor Challenge").
Title: Name of the Team. Then add all participants with their emails.

B. Team Composition

There is no limit to team size. Only supervisors can participate in multiple teams within the same challenge.

C. Code and Data Sharing

Private sharing of code or data outside your team is strictly prohibited.

D. Challenge Completion Requirements

To successfully complete the challenge, participants must provide the submission information as per the deadlines.

Evaluation Guidelines

Recognition (F1-Score, Accuracy): Submissions should demonstrate high accuracy and F1-scores

Paper Submission

The paper should be written in English, following the formatting guidelines. The paper can be 8-12 pages including references.

Prizes

All participating teams will receive a Certificate of Participation. Winning Team(s) may get surprising prizes!

Dataset Information

The dataset for the challenge has been collected using using various sensors and mobile applications to ensure a comprehensive multimodal approach to activity recognition and classification. However, we provide a partial data for you. FonLog (Japan): A mobile application that logs sensor data and tracks user activities, providing rich contextual information for activities in real-world settings.

Data Structure

Sensor Data

Main Directory: users_timeXYZ/users/

  • This directory contains several subdirectories, each named with random numbers (e.g., '38', '1716').
  • Each subdirectory contains one or more CSV files containing accelerometer sensor data.
  • The CSV file naming convention follows the pattern: user-acc_[DIR-NUMBER]_[TIMESTAMP]_[RANDOM-NUMBER].csv.

Example: Folder '38' contains files such as:

  • user-acc_38_2024-09-08T23_31_01.510+0100_97016.csv
  • user-acc_38_2024-09-08T23_31_16.519+0100_15638.csv

Sensor Data File Format

Each CSV file consists of five columns:

  • Random identifier (can be ignored)
  • Timestamp
  • X-axis accelerometer reading
  • Y-axis accelerometer reading
  • Z-axis accelerometer reading

Activity Labels

The dataset captures a range of activity classes that include both normal and unusual activities associated with Parkinson’s disease. These classes are designed to help identify and differentiate patterns of movement and behaviors.
  • (Facing the camera) Sitting and standing
  • (FACING camera) both hands SHAKING (sitting position)
  • Stand up from chair - both hands with SHAKING
  • (Sideway) Sit & stand
  • (Sideway) both hands SHAKING (sitting)
  • (Sideway) STAND up with - both hands SHAKING
  • Cool down - sitting/relax
  • Walk (LEFT --> Right --> Left)
  • Walk & STOP/frozen, full body shaking, rotate then return back
  • Slow walk (SHAKING hands/body, tiny step, head forward)
Relax Time: Represents periods of rest or minimal physical activity.
Normal Sit and Stand: Routine activities of sitting down and standing up, performed without motor abnormalities.
Tremor While Seated: Involuntary and rhythmic shaking of body parts, such as hands or legs, occurring while the subject is seated. Indicates motor impairments specific to Parkinson’s disease.
Tremor While Standing: Similar to seated tremors, this involves involuntary movements but occurs while the subject is in a standing position. Provides data on balance and postural stability under motor impairment conditions.
Normal Walking: Regular walking pattern without disruptions or irregularities. Acts as a baseline for comparing other gait-related abnormalities.
Freezing and Festinating Gait: Freezing Gait: Episodes where movement temporarily halts, making it difficult to initiate or continue walking. Festinating Gait: Accelerated, short steps often leading to instability. Both are critical indicators of advanced Parkinson’s disease.
Shuffling Gait: A distinctive walking pattern characterized by dragging feet or taking short steps. Common in individuals with Parkinson’s disease and is a marker of progression.

File: TrainActivities.csv: Contains 7 columns:

  • ID (random identifier)
  • Activity Type ID
  • Activity Type (10 distinct activity classes): You need to recognize them.
  • Start Time
  • End Time
  • Update Time
  • Subject ID (e.g., U1, U2, U3, ..., U21, U22)

Training Data

  • 9 subjects provided for training. 9 subjects are: U1, U2, U3, U4, U5, U6, U7, U21, and U22.
  • Each subject performs all 10 activities unless any missing data.
  • Most activities have multiple repetitions. Note: Some time gaps may exist due to data collection issues

Important Notes

  • Objective: Develop a model to recognize 10 different activities based on accelerometer data. Link accelerometer data with corresponding activities and subjects. Develop recognition model(s) for the 10 activity classes. Implement appropriate train/test splitting strategies. Document your approach and results so that you can write a paper.
  • The user IDs, random numbers, and file names are anonymised and randomly assigned for privacy protection.
  • There could be missing data and inconsistencies due to real-world conditions during data collection.
  • Some users have multiple activity sessions recorded, and timestamps may overlap.
  • Significance: This dataset is uniquely positioned to offer insights into the daily activities of individuals with Parkinson's disease by leveraging multimodal data sources. The combination of wearable devices, skeleton tracking, and application-based logging ensures diverse and high-quality data for understanding normal and unusual activity patterns. By integrating data from different sensors, this dataset provides a valuable resource for developing and testing machine learning models that can identify and classify activities with potential applications in disease monitoring and patient care.
  • Data Sharing Policy: The dataset will be provided exclusively to registered participating teams and is strictly limited to use for the challenge.

Tutorial

Data preprocessing code is NOT PROVIDED, as it is intended to be part of the challenge. Since the data provided is not particularly complex, the data analysis itself becomes part of the task. Therefore, the preprocessing code will not be provided.

Slide: PPT slide.

FAQ

Frequently Asked Questions (FAQ)

Q1. Missing Data in TrainActivities Dataset - many time periods do not have corresponding accelerometer data in users_timeXYZ. Is this expected or an issue with the dataset?

Indeed, the missing data in the accelerometer readings is unintentional and reflects a common characteristic of real-world data collection. Such gaps can occur due to various reasons, including sensor malfunctions, user movement out of range, or environmental interference.

One of the core objectives of this challenge is to simulate the complexities of real-world scenarios, encouraging participants to develop innovative techniques to handle incomplete 'in-the-wild' datasets effectively. This is a crucial aspect of deploying robust solutions for human activity recognition in practical settings. If you need further assistance or have additional questions, feel free to reach out. We are excited to see how you and other participants approach these challenges creatively.

Q2. Is there a difference in timezone on sensor and labels?

The timezones for the sensor data and label data are indeed different. While the timezone for the sensor data can be determined from the timestamps, we noticed that the timestamps in the label data are not properly annotated. Therefore, please consider that the label data is based on Japan Standard Time (JST), which is UTC+09:00.

Q3. Are the timestamps in order?

The timestamps are in sequence. Please handle the timestamps by keeping this information in mind alongside the difference in timezone.

Q4. Are there multiple activities happening in the same minute?

Some activities are completed within a few seconds and are performed quickly one after another, in order to replicate natural work settings. Others may span several minutes. Thus, within a one-minute timeframe, there may be one or multiple activities.

Q5. How do we submit the test data?

Participants are required to submit the output CSV file, ensuring that it includes the activity label for each corresponding feature/row. The test file should remain unchanged, except for the addition of one extra column named "OUTPUT" containing the predicted label.

You will have 164 rows in the test data "TestActivities-20240920.csv". You must retain the values of the test data exactly and simply add the new "OUTPUT" column at the end, providing one label per row.

Q6. Regarding the tutorial, is there a Google Colab notebook to be released?

Data preprocessing code is not provided, as it is intentionally made part of the challenge. Since the data is not particularly complex, analysing and handling the data independently forms an essential element of the task. Therefore, the preprocessing code will not be shared.

You are encouraged to visit the Tutorial Page to learn more about the challenge.

Q7. Access to Dataset - Where can we get the data?

The dataset will be provided exclusively to registered participating teams and is strictly limited for use in the challenge. After registering your team, the host will email you the data. Kindly review the Dataset Page to learn more about the dataset and the Data Sharing Policy.

Q8. How can we register our team?

All participants, including solo challengers, must complete and submit the Team Registration Form prior to joining the competition. In the form, please fill in the following fields:

  • Title: Enter your team name.
  • Abstract: This field may be left blank.
  • Track Selection: Choose the specific challenge track you are participating in.

To know more about the registration guidelines and requirements, kindly visit the Participation Page.

Q9. Where can we submit the paper and results?

You can submit your challenge paper via CMT.

Q10. Can we reuse the data for research purposes?

The data is intended for challenge use only. Should you wish to continue the analysis for research purposes, kindly contact the challenge hosts from the University of East London (Professor Md Atiqur Rahman Ahad or Dr Shahera Hossain).

Q11. What is the format for paper submission?

Challenge papers follow the IJABC Journal format as Overleaf Template. Ideally, submissions should be approximately 10 pages in journal format, including references. Although IJABC has a flexible length policy, a minimum of 10 pages would be reasonable to adequately explain your approach with supporting related work.

Organizers:

Prof. Dr. Md Atiqur Rahman Ahad, University of East London, UK
Dr. Shahera Hossain, University of East London, UK
Dr. Tahera Hossain, Nagoya University, Japan
Dr. Christina Garcia, Kyushu Institute of Technology, Japan
Md. Mamun Ibrahim, University of East London, UK
Tahia Tazin, UNSW, Australia
Prof. Dr. Sozo Inoue, Kyushu Institute of Technology, Japan