The code and dataset for our paper "Weakly-supervised outlier detection in mixed-attribute data with fuzzy rough sets".
We use 16 public datasets to assess the model performances, including 2 nominal, 2 mixed, and 12 numerical datasets. The details of the datasets are provided in below table:
| No | Datasets | # Objects | # Attributes | # Outlier | % Outlier | Category | Data Type |
|---|---|---|---|---|---|---|---|
| 1 | annthyroid | 7200 | 6 | 534 | 7.4% | Healthcare | Numerical |
| 2 | Arrhythmia | 452 | 279 | 66 | 14.6% | Medical | Mixed |
| 3 | Cardiotocography | 2114 | 21 | 466 | 22.0% | Healthcare | Numerical |
| 4 | Ionosphere | 351 | 32 | 126 | 35.9% | Oryctognosy | Numerical |
| 5 | mammography | 11183 | 6 | 260 | 2.3% | Healthcare | Numerical |
| 6 | Mushroom1 | 4429 | 22 | 221 | 5.0% | Botany | Categorical |
| 7 | Mushroom2 | 4781 | 22 | 573 | 12.0% | Botany | Categorical |
| 8 | musk | 3062 | 166 | 97 | 3.2% | Chemistry | Numerical |
| 9 | optdigits | 5216 | 64 | 150 | 2.9% | Image | Numerical |
| 10 | PageBlocks | 5393 | 10 | 510 | 9.5% | Document | Numerical |
| 11 | Pima | 768 | 8 | 268 | 34.9% | Healthcare | Numerical |
| 12 | satellite | 6435 | 36 | 2036 | 31.6% | Astronautics | Numerical |
| 13 | satimage-2 | 5803 | 36 | 71 | 1.2% | Astronautics | Numerical |
| 14 | Sick | 3613 | 29 | 72 | 2.0% | Medical | Mixed |
| 15 | SpamBase | 4207 | 57 | 1679 | 39.9% | Document | Numerical |
| 16 | thyroid | 3772 | 6 | 93 | 2.5% | Healthcare | Numerical |
- cudatoolkit=11.6.0
- numpy=1.23.5
- pandas=1.5.3
- python=3.8.16
- pytorch=1.12.1
- scikit-learn=1.2.0
- scipy=1.9.3
- torchaudio=0.12.1
- torchvision=0.13.1