Table 2 Characteristics and descriptions of datasets for classification from UCI repository used for hyperparameter sensitivity analysis and for comparisons
From: Extreme learning machine versus classical feedforward network
Dataset | No of classes | No of samples | No of features | Minority class size | Majority class size | Description |
|---|---|---|---|---|---|---|
Breast Cancer (Ljubljana) | 2 | 277 | 41 | 81 | 196 | Prediction of a breast cancer recurrence. All features are one-hot encoded |
Breast Cancer (Wisconsin) | 2 | 569 | 30 | 212 | 357 | Prediction of a breast cancer class (benign or malignant) [28] |
CNAE-9 | 9 | 1080 | 856 | 120 | 120 | Documents represented with bag-of-words; sparse and high-dimensional encoding, classes are equinumerous |
Forest Cover Type | 7 | 581,000 | 54 | 2747 | 283,000 | Forest classification based on cartographic features; large dataset with imbalanced classes |
Glass Identification | 6 | 214 | 9 | 9 | 76 | Classification of glass shards collected as crime evidence. Highly imbalanced classes |
Landsat | 6 | 6435 | 36 | 626 | 1533 | Land cover classification using satellite image continuous features. Classes are imbalanced |
Sonar | 2 | 208 | 60 | 97 | 111 | Land mines and stones classification based on a sonar spectrogram; high dimensionality-to-size ratio, classes balanced |
Urban Land Cover | 9 | 675 | 147 | 29 | 122 | Land cover recognition using features from aerial images; relatively small, highly dimensional |
Wine | 3 | 178 | 13 | 48 | 71 | Classification of 3 types of wine |