Usage

Check the file u_shapelet.ipynb for tutorial.

Rerun the experiment using the file run_in_parallel.py.

The datasets: uncertain-dataset.tar.gz

Results

• The accuracy of each considered method and each uncertainty level is here: all_results.csv

Critical difference diagrams of models accuracy rank(lower is better)

| | Low uncertainty | Medium uncertainty | High uncertainty | | ---- | ------------------------------------------------------------ | ------------------------------------------------------------ | ------------------------------------------------------------ | | NB | | | | | RF | | | | | All | | | |

Accuracy scatter plots of UST(UED,RF) vs others

• Low uncertainty

  

• Medium uncertainty

  

• High uncertainty

  

Critical difference diagrams of models log loss (lower is better)

| | Low uncertainty | Medium uncertainty | High uncertainty | | ---- | ------------------------------------------------------------ | ------------------------------------------------------------ | ------------------------------------------------------------ | | NB | | | | | RF | | | | | All | | | |

Dependencies

• imbalanced-learn=0.7.0
• numpy==1.19.5
• pandas==1.2.0
• sktime==0.5.1

Cite this work

@INPROCEEDINGS{mbouopda@2020,  
  author={Mbouopda, Michael Franklin and Mephu Nguifo, Engelbert},  
  booktitle={2020 International Conference on Data Mining Workshops (ICDMW)},   
  title={Uncertain Time Series Classification with Shapelet Transform},   
  year={2020},  
  volume={},  
  number={},  
  pages={259-266},  
  doi={10.1109/ICDMW51313.2020.00044}
}