A Spacecraft Electrical Characteristics Multi-Label Classification Method Based on Off-Line FCM Clustering and On-Line WPSVM
Fig 9
The sample that has high membership can be confirmed directly; other samples which belong to multiple categories with similar membership require further revision by human experts. Compared with k-means algorithm, the FCM algorithm offered an extra matrix U which uses a threshold set by us. Only the elements of the matrix U which are less than the threshold need to be corrected by humans. The matrix U will save much human error correction time in the off-line system.