4 Experiments
This experiment is implemented on IDEA Java platform with the help of Weka API.
Concretely, Java compiler environment is JDK 8 and data mining software is Weka 3.9
[8]. The REPTree in Weka is selected as the typical classification algorithm for
comparison and the basic classification algorithm of ensemble learning. In the experiments, REPTree adopts the post pruning strategy [9] and the branching criterion based
on information gain. The number of base classifiers of the Bagging algorithm is 50, and
the other parameters are set by default in Weka.
In this section, we present four experiments. First, the comparison experiment of
Bagging and REPTree under ideal conditions. Second, the comparison experiment
under the influence of noise. Third, the comparison experiment under the influence of
small sample set. Finally, the comparison experiment in consideration of rejection. The
data set of the decoy jamming effect generated by simulation is used as experimental
data. For each case, five times of two-fold cross-validation experiments were carried
out, i.e., two-fold cross-validation was repeated five times. During two-fold crossvalidation, the original data set is randomly divided into two equal parts as training set
TR and test set TE. TE is invisible in training process, and it is specially used to
evaluate the accuracy of classification algorithm.
4.1 Comparison Experiment Under Ideal Conditions
Ideal conditions represent that there is no object outside the library and it is not affected
by noise sample and small sample set. The confusion matrix is used to study the
separability between targets. The concrete result of the classification algorithm can be
calculated from the confusion matrix. The confusion matrices of the two classification
algorithms are shown in Tables 1 and 2 respectively.
Table 1. The confusion matrix of REPTree under ideal conditions (%)
True
Predicted
−1
1
−1
99.33
0.66
1
0.17
99.83
Table 2. The confusion matrix of Bagging under ideal conditions (%)
True
Predicted
−1
1
−1
99.45
0.55
1
0.13
99.87
292
X. Wei and X. Fan
This experiment is implemented on IDEA Java platform with the help of Weka API.
Concretely, Java compiler environment is JDK 8 and data mining software is Weka 3.9
[8]. The REPTree in Weka is selected as the typical classification algorithm for
comparison and the basic classification algorithm of ensemble learning. In the experiments, REPTree adopts the post pruning strategy [9] and the branching criterion based
on information gain. The number of base classifiers of the Bagging algorithm is 50, and
the other parameters are set by default in Weka.
In this section, we present four experiments. First, the comparison experiment of
Bagging and REPTree under ideal conditions. Second, the comparison experiment
under the influence of noise. Third, the comparison experiment under the influence of
small sample set. Finally, the comparison experiment in consideration of rejection. The
data set of the decoy jamming effect generated by simulation is used as experimental
data. For each case, five times of two-fold cross-validation experiments were carried
out, i.e., two-fold cross-validation was repeated five times. During two-fold crossvalidation, the original data set is randomly divided into two equal parts as training set
TR and test set TE. TE is invisible in training process, and it is specially used to
evaluate the accuracy of classification algorithm.
4.1 Comparison Experiment Under Ideal Conditions
Ideal conditions represent that there is no object outside the library and it is not affected
by noise sample and small sample set. The confusion matrix is used to study the
separability between targets. The concrete result of the classification algorithm can be
calculated from the confusion matrix. The confusion matrices of the two classification
algorithms are shown in Tables 1 and 2 respectively.
Table 1. The confusion matrix of REPTree under ideal conditions (%)
True
Predicted
−1
1
−1
99.33
0.66
1
0.17
99.83
Table 2. The confusion matrix of Bagging under ideal conditions (%)
True
Predicted
−1
1
−1
99.45
0.55
1
0.13
99.87
292
X. Wei and X. Fan