Page 522 - Decision Making Applications in Modern Power Systems

A reliable decision-making algorithm Chapter | 18 481

Input Bootstrapping Bagging Aggregation

Sample (1) Decision tree-1

Ensembler
Dataset Sample (2) Decision tree-2
(Avg/Vote)
Sample (N) Decision tree-N

FIGURE 18.5 Architecture of a BDT model. BDT, Bagged decision tree.

such tree-based ensemble technique is BDT that builds, by randomization,
numerous DTs and then aggregates their predictions as exemplified in
Fig. 18.5. From an ensemble of trees, one can derive an importance score for
each variable of the problem that assesses its relevance for predicting the out-
put. Every tree in the ensemble is grown on an independently drawn bootstrap
replica of input data [29]. Observations not included in this replica are “out-
of-bag” for this tree [30]. The bagging is nothing but a bootstrap aggregation
of a set of DTs. Individual DTs tend to over-fit. BDT (Bootstrap aggregated)
combines the results of many DTs, thereby overwhelming the effects of
over-fitting and improves generalization. Tree Bagger grows the DTs in the
ensemble using bootstrap samples of the data. The estimated out-of-bag error
variances can be reduced by setting a more balanced misclassification cost
matrix or a less skewed prior probability vector. BDT generates in-bag
samples by oversampling classes with large misclassification costs and under-
sampling classes with small misclassification costs. Consequently, out-of-bag
samples have fewer observations from classes with large misclassification
costs and more observations from classes with small misclassification costs.
To train a BDT ensemble classifier using a small data set, a highly skewed
cost matrix is considered, and then the number of out-of-bag observations per
class might be very low. In both bagging and random forest, many individual
DTs are built on a bootstrapped version of the original dataset and are ensem-
ble together. The difference is that, in a random forest, a random subset of
variables are considered during node split while in bagging of DTs, all the
variables are considered in a node split [31].

18.4.3 Wavelet packet energy and bagged decision tree based
decision-making algorithm

In order to train and test the proposed WPE_BDT based decision-making
algorithm for detection and classification of shunt faults during PS condition,