Skip to main content

Table 2 The results over imbalanced, trimmed and balanced data.

From: Prediction of protein-protein interaction sites using an ensemble method

Features Dataset AUC Sensitivity Specificity
Group 1 Imbalanced 0.71815 0.23618 0.95589
  Trimmed 0.69259 0.60103 0.67216
  Balanced 0.72944 0.62285 0.70575
Group 2 Imbalanced 0.74139 0.26960 0.96100
  Trimmed 0.74575 0.73217 0.62533
  Balanced 0.77802 0.69670 0.71086
Group 3 Imbalanced 0.81745 0.37526 0.95015
  Trimmed 0.81670 0.72653 0.74426
  Balanced 0.84647 0.76836 0.76798
Group 4 Imbalanced 0.80099 0.27180 0.96929
  Trimmed 0.79362 0.72478 0.71542
  Balanced 0.83079 0.73978 0.75002
  1. The imbalanced data includes all examples in the original datasets; the trimmed data owns all positive examples and randomly selected negative examples, with a 1:1 ratio of positive to negative examples; the balanced datasets are generated by Sub-EnClassifiers with resampling technique.