TSCGlue: A Robust and Efficient Stacking Framework for Time Series Classification (anon)

zenodo_results.parquet contains classification accuracy results for time series classification models evaluated on 112 UCR datasets using 30 stratified resamples each. Each row corresponds to one model–dataset–fold combination and includes the test accuracy. Results were produced using the experimental scripts in this repository and collected from individual per-fold runs. zenodo_timing.parquet contains train and inference timing benchmarks for the TSCGlue model aggregated across folds. Each row corresponds to one dataset and includes mean train time (seconds), mean predict time (seconds), training set size, test set size, and series length.