128 time-series classification datasets updated from the 2018 UCR Archive to address data integrity issues. The collection includes specific fixes for backward compatibility, handling of missing values, and support for varying sequence lengths across diverse domains like medicine, robotics, and sensor monitoring.
Use Cases
- Benchmark time-series classification algorithms using the provided train/test splits and class labels
- Evaluate the performance of Dynamic Time Warping (DTW) on datasets specifically marked with varying lengths
- Test imputation techniques on sequences containing missing values to improve classification accuracy
- Compare model performance against established state-of-the-art results for the 128 included datasets
Strengths
- 128 time-series classification datasets covering domains from ECG to motion capture
- Standardized handling for datasets with varying sequence lengths
- Resolved missing value entries within the time-series data points
- Maintains backward compatibility with the original 2018 UCR Archive release