Page 488 - Decision Making Applications in Modern Power Systems

448 Decision Making Applications in Modern Power Systems

FIGURE 17.2 The details of feature vector construction.

features may be extracted from the sampled signals at the relay location. The
procedure of feature selection selects a subset from the original features. The
feature selection is performed to

select informative and pertinent features and reduce the feature set,
save memory space and decrease computation burden of the algorithm,
improve the classification and/or predictive accuracy,
enhance generalization ability, and
represent knowledge about the extracted features.
In this chapter the feature selection techniques using supervised models
are investigated. The techniques are classified into filter, wrapper, and
embedded [18].
Filter techniques rank features to filter out the less relevant variables before
classification. How to measure the relevancy of a feature to the other features
or the data or the target output is one of the main challenges of filter techni-
ques. A proper ranking criterion is employed to score the input features, and a
feature that has the score below the threshold can be a candidate to remove.
Filter techniques are more useful to represent the relationships between the
input features, whereas they don’t consider the performance of a classifier or
predictor model. Many filter techniques achieve only a feature ranking based
on the measuring of relationships instead of the best subset of input feature.
Mutual information-based methods, correlation coefficient based criteria,
relief-based algorithms are some of the examples of filter techniques.