Datasets and partitions used in the paper:
Weight Selection Strategies for Orderd Weighted Average Based Fuzzy Rough Sets
S. Vluymans, N. Mac Parthaláin, C. Cornelis, Y. Saeys
Submitted to Information Sciences
Corresponding author: sarah.vluymans@ugent.be
In this paper, our proposed OWA weight selection guidelines were validated on 20 independent datasets, taken from the KEEL, UCI and WEKA dataset repositories. A zip-file containing these datasets and the partitions used in the paper is available for download: OWA_Weight_Selection_datasets.zip.
Their characteristics are described in the table below.
For each dataset, we list the number of features (nFeat), number of instances (nInst), number of classes (nCl) and the imbalance ratio (IR). We also specify whether or not all features are nominal.
| Name | nFeat | All nominal? | nInst | nCl | IR |
|---|---|---|---|---|---|
| appendicitis | 7 | No | 106 | 2 | 4.05 |
| banknote | 7 | No | 1372 | 2 | 1.25 |
| biodeg | 40 | No | 1055 | 2 | 1.96 |
| credit | 15 | No | 653 | 2 | 1.21 |
| ctg | 21 | No | 2126 | 10 | 10.92 |
| eye_detection | 14 | No | 14980 | 2 | 1.23 |
| faults | 32 | No | 1941 | 2 | 1.88 |
| grub | 8 | No | 155 | 4 | 2.58 |
| hepatitis | 19 | No | 80 | 2 | 5.15 |
| housevotes | 16 | Yes | 232 | 2 | 1.15 |
| iris | 4 | No | 150 | 3 | 1 |
| letter | 16 | No | 20000 | 26 | 1.11 |
| magic | 10 | No | 19020 | 26 | 1.11 |
| messidor | 19 | No | 1151 | 2 | 1.13 |
| mushroom | 22 | Yes | 5644 | 2 | 1.62 |
| optdigits | 64 | No | 5620 | 10 | 1.03 |
| penbased | 16 | No | 10992 | 10 | 1.08 |
| seismic | 18 | No | 2584 | 2 | 14.2 |
| sensor | 24 | No | 5456 | 4 | 6.72 |
| transfusion | 4 | No | 748 | 2 | 3.2 |
The full results for the FRPS and POSNN methods, discussed in Section 5.4 of the paper are available: OWA_Weight_Selection_FRPS_and_POSNN.xlsx.