Correcting the Training Data

Restricted Decontamination for the Imbalanced Training Sample Problem

“…In a previous work [2], a methodology for correcting a TS while employing nonparametric classifiers has been presented. The Decontamination procedure can be regarded as a cleaning process removing some elements of the TS and correcting the label of several others while retaining them.…”

“…In the present paper, we introduce a new proposal for balancing the TS through reduction of the majority class size and, at the same time, an increase in the amount of prototypes in the minority class. To this aim, we employ a modification of the Decontamination methodology [2] that will be referred to as Restricted Decontamination. We also explore the convenience of using this technique in combination with a weighted distance measure aimed at biasing the classification procedure.…”

Improving the k-NN method: Rough Set in edit training set

“…The Generalized Editing Algorithm consists of removing some suspicious prototypes and changing the class labels of some other instances. Accordingly, it can be regarded as a technique for modifying the structure of the training sample (through re-labeling of some training instances) and not only for eliminating atypical instances [20]. The second algorithm is proposed taking into account these ideas.…”

“…This technique consists in applying the k-NN (k > 1) classifier to estimate the class label of every prototype in the training set and discard those instances whose class label does not agree with the class associated to the majority of the k neighbors. The benefitsimprovements of the generalization accuracy-of Wilson's algorithm have been supported by theoretical and empirical evaluations [20]. The Repeated ENN (RENN) applies the ENN algorithm repeatedly until all instances remaining have a majority of their neighbors with the same class, which continues to widen the gap between classes and smoothes the decision boundary.…”

“…Editing algorithms from the training sample are described in [20], which are focused on the detection and elimination of noisy or atypical patterns in order to improve the classification's exactitude. Some of these are ENN (Wilson, 1972), All k-NN (Tomek, 1976) and Generalized Editing Algorithm (Koplowitz and Brown, 1978).…”