*5.2. Extracted Feature Sequence by Unsupervised MI V.S. Supervised MI*

Table 3 shows the comparison of feature sequences extracted by the unsupervised mutual information method and supervised mutual information method and their corresponding weights. As obtained from Table 3.

Table 3 Extracted feature sequence by unsupervised MI V.S. supervised MI the mutual information of unsupervised feature extraction method selects the same first feature from cases of nuclear power transformer DGA with that of supervised method, while the weight value, generated by supervised feature extraction algorithm is greater than the corresponding values in the unsupervised algorithm, which means supervised method has more dynamic to choose the first feature; In other steps, selected feature is not the same, but the unsupervised algorithm has a strong dynamic at each step.
