Credal Dog-4

Most datasets used for classification use hard labels. Credal Dog-4 was labeled uncertainly and imprecisely by contributors during crowdsourcing campaigns. Resulting in richer labels, modeled with the theory of belief funtions, which generalizes several reasoning frameworks with uncertainty. These datasets can be used with classical models using hard labels but also with probabilistic, fuzzy or even evidential models.

Dataset:
4 classes, 400 observations, 47 features (or raw pictures)

When using the dataset please cite :
A. Hoarau, C. Thierry, A. Martin, J.-C. Dubois, Y. Le Gall, "Datasets with rich labels for machine learning", in: FUZZ-IEEE, 2023.

Credal Dog-4 dataset
├── data
│ ├── classes.csv: Classes of the dataset
│ ├── X.csv: Features of the dataset
│ ├── X_512.csv: Learge 512 features vector of the dataset
│ ├── X_pictures.csv: Raw features (Pictures themselves)
│ ├── y.csv: Rich labels
│ └── y_true.csv: True labels
└── extra
├── y_hard.csv: Hard labels given during a new campaign
├── DATA_imperfect.csv: Imperfect answers given during the campaign
├── ITERATION_imperfect.csv: Imperfect 2nd step answers given during the campaign
├── EVENT_imperfect.csv: Contributors events
├── ID_imperfect.csv: Contributors IDs
└── DATA_perfect.csv: Precise answers given during a new campaign

Credal Dog-7:
https://data.mendeley.com/datasets/4hz3wx6wm5
Credal Dog-2:
https://data.mendeley.com/datasets/95mmjfrsyh
Credal Bird-10:
https://data.mendeley.com/datasets/9sz5vm8g7h
Credal Bird-2:
https://data.mendeley.com/datasets/v8k6c5s9cy