“…This approximate way of dealing with EDCs suffers the drawbacks of sampling techniques, which can modify the problem by reducing the influence of critical samples and/or emphasizing unimportant instances [20]. Decision trees have also been considered for EDC problems [21], [22], and perceptrons and piecewise linear classifiers were used in [23] with an hybrid learning algorithm that constructs separating hyperplanes for each pair of classes. In general, all these techniques suffer from the limitations of the constrained partition of the input space.…”