Hanczar08a
Supplementary Information for:
Classification with reject option in gene expression data
Blaise Hanczar, Edward R. Dougherty
Experimental Design:
Details on the exprimental design for simulations based on artificial and microarray data.
- Experiments with artificial data (fig. C1, C2)
- Experiments with microarray data (fig. C3)
Supplementary results:
The performance of classification with rejection option on microarray dataset. The results are computed by 10-fold cross-validation.
- Supplementary results on breast cancer dataset (fig. C4)
- Supplementary results on lung cancer dataset (fig. C5)
- Supplementary results on colon cancer dataset (fig. C6)
Comparison with classifiers using posterior probabilities
We compare our method with the classifiers using the posterior probabilities. This kind of classifier has fixed thresholds, unlike to our method that use adaptive thresholds. These classifiers reject an examples if the highest posterior probability is lower than a given threshold. This simulations have been done on artificial datasets. These artificial data have been generated with Gaussian mixture models lernt from microarray data.
- Results on artificial data based on breast cancer dataset (fig. C7, C8)
- Results on artificial data based on lung cancer dataset (fig. C9, C10)
- Results on artificial data based on colon cancer dataset (fig. C11, C12)
Baseline results
Results on a "null dataset". There is not relation between the features and the class.
- Results (fig. C13)
