Skip to main content

Table 3 This table shows the main results comparing the feature selection benchmarks of our approach with Lasso, 1-SVM, and Maldi-Quant. These are averages over 10 repetitions of a 5-fold cross-validation. Note that these results have been calculated based on the highest accuracy criterion for all classifiers with between 10 and 30 selected features. This particularly means that better accuracy values might be achieved for the individual methods if less sparse feature vectors would be allowed. For more details see text

From: Sparse Proteomics Analysis – a compressed sensing-based approach for feature selection and classification of high-dimensional proteomics mass spectrometry data

  SPA Lasso 1-SVM Maldi-Quant
Dataset Feat. [e] Sens [f] Spec [g] B. Acc [h] Feat. Sens Spec B. Acc Feat. Sens Spec B. Acc Feat Sens Spec B. Acc
P. CA - UHL 20.48 0.975 0.949 0.962 29.94 0.969 0.939 0.954 27.72 0.964 0.947 0.955 21 0.888 0.888 0.888
P. CA - UHh 17.1 0.986 0.975 0.981 26.46 0.966 0.969 0.967 29.68 0.966 0.976 0.971 17 0.975 0.963 0.969
  1. [e]Feat.: Number of features
  2. [f]Sens: Sensitivity (T P/(T P+F N))
  3. [g]Spec: Specificity (T N/(F P+T N))
  4. [h]B. Acc: Balanced Accuracy (\( \frac {\text {sens.} + \text {spec.}}{2}\))