MetaPassAV: Passive Behavior Tracking of Avatars

An anonymized behavioral authentication dataset based on **passive avatar** kinematics in a **metaverse** environment.

Overview

This repository provides an anonymized behavioral authentication dataset collected in a custom virtual art gallery using Meta Quest 2. This timeseries data was collected by tracking the passive behavioral kinematics of avatars (AVs) in custom immersive metaverse environment. The participants were seated and allowed to freely explore the virtual gallery without any assigned task. The dataset is intended for research in behavioral authentication, user identification, and related fields.

Dataset Summary

Property	Value
Participants	16
Sessions	123
Total Duration	4 h 53 min
Average Session Length	2 min 23 sec
Sampling Rate	0.25 Hz
Raw Features	14
Engineered Features	28
Sliding Windows	16,291
Window Length	44 s
Window Stride	2 s

Collection

• Custom Unity virtual art gallery
• Meta Quest 2 headset
• Seated interaction
• Passive exploration (no assigned task)
• Navigation via controllers
• Natural head movement recorded
• One sample every 4 seconds plus image-view events

Repository

.
├── README.md
├── MetaPassAV-v1.pdf
├── DataPrep/
|   ├── 1 Data Preparation.ipynb
|   ├── 2 EDA and Split.ipynb
|   ├── data/
|   └── scripts/
└── UnifiedAVDTAuth/
    ├── README.md
    └── unifiedAuthFramework/
        ├── 01_KAN_AE.ipynb
        ├── 02_BiGRU_Classifier.ipynb
        └── 03_UBM_Template_Auth.ipynb

Steps to Reproduce

1. Clone the repository
2. Navigate to the DataPrep folder
3. Run the notebooks in order:
   • 1 Data Preparation.ipynb
   • 2 EDA and Split.ipynb
4. The exported data will be available in the DataPrep/data/raw_split folder.

The final dataset is available in the DataPrep/data/raw_split folder as a CSV file. The dataset contains 14 raw features and 28 engineered features, with a total of 16,291 sliding windows of 44 seconds each, sampled every 2 seconds. It also consits of 2 identifiers (UserID and SampleID) for each sliding window, make complete column count as 30.

The final dataset can be used for developoing the proposed template-based deep-UBM inspired behavioral authentication model, as described in the paper (DOI). The core code and implementation of the model is available in the UnifiedAVDTAuth directory.

Citation

While using this dataset or the proposed model, please cite the following accompanying publication titled “A Unified Framework for Behavior-Based Identification and Authentication of Avatars and Digital Twins”:

@article{PRAKASH2026116714,
 title={A unified framework for behavior-based identification and authentication of avatars and digital twins},
 journal={Applied Soft Computing},
 pages={116714},
 year={2026},
 issn={1568-4946},
 doi={https://doi.org/10.1016/j.asoc.2026.116714},
 url={https://www.sciencedirect.com/science/article/pii/S1568494626021629},
 author={Ravi Prakash and Tony Thomas},
}

For detailed reference for this dataset and code repository, the citation can be found in the CITATION.cff file.

License

This dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) License.

You are free to share and adapt the dataset for any purpose, provided appropriate attribution is given.

See the LICENSE file for complete details.