WhichFinger Dataset from CAFO: Feature-Centric Explanation on Time Series Classification

This is the WhichFinger Multivariate Time Series (MTS) dataset from the SIGKDD 2024 paper CAFO: Feature-Centric Explanation on Time Series Classification by Jaeho Kim, Seok-Ju (Adam) Hahn, Yoontae Hwang, Junghye Lee, and Seulki Lee. Basically, we ask participants to repeat (a) and (b) from the figures for one minute (it looks easy, but its tough!).

The WhichFinger Dataset For Time Series Explanation.

The WhichFinger Dataset is a multivariate time series (MTS) dataset, designed for eXplainable Artificial Intelligence (XAI) applications. This dataset offers comprehensive information on the data collection process for each class, as well as the features relevant to specific classes, which facilitates the validation of the CWRI measure. We created this dataset because, to the best of our knowledge, no public MTS datasets met the following three criteria: (1) strong prior knowledge or information regarding each feature's contribution to specific classes, (2) a sufficient number of classes $(C\geq2)$ and features $(D\geq2)$, and (3) an adequate number of samples $(N\geq1,000)$. In this section, we describe the detailed data collection process and the preprocessing steps involved in the creation of the dataset.

Please kindly refer to Appendix G: WhichFinger Dataset of our paper for further details.

Simple Statistics.

• Total Sample Size: 18,010
• Windowed Time Series Length: 120 (You can change this as we provide the raw data)
• Users: 19
• Features: 10 Sensors
• Frequency: 66.7 Hz
• Class: (1) Thumb only, (2) Thumb except, (3) Index only, (4) Index except, (5) Middle only, (6) Middle except, (7) Ring only, (8) Ring except, (9) Pinky only, and (10) Pinky except

GoogleDrive Link

https://drive.google.com/drive/folders/16Xks-9O6BeTFHOba9-HZRZcnd2GnrUVg?usp=sharing

GoogleDrive Contents

• WhichFinger/raw_data_df.pkl: This contains the whole raw data. You can load this file in the notebook/preprocess.ipynb for preprocessing, or you can use the below.
• WhichFinger/WhichFinger_ModelTraining: Containing the preprocessed files used for model training.
• answer_sheet.csv: The Answer Sheet used to evaluate the CWRI metrics.
• finger_dataset.py: A PyTorch Dataset class
• fingergesture.pkl: We used this file for model training. It is basically the same file.
• label_df.csv: Contains meta info, and y_true.

Contents

• notebook/01_preprocess_raw.ipynb: This notebook contains the preprocessing script for the WhichFinger.
• notebook/02_evaluate_cwri.ipynb: This notebook contains the minimal code to evaluate our CWRI score.

Ground Truth Evaluation

Please kindly refer to Appendix E: Evaluation of CWRI Metrics for further explanation.

License for Data Use

We apply the Creative Commons Attribution-NonCommercial 4.0 International License.

Special Thanks

We are grateful to Prof. Sunghoon Lim, Gyeongho Kim, Sujin Jeon, and Jae Gyeong Choi for their invaluable contributions to our research. Their provision of the smart glove was essential for the WhichFinger data collection. We also thank the 20 participants. For more information about the smart glove, visit FTSAME.

Citation