Similar component analysis

A new unsupervised feature extraction method called similar component analysis (SCA) is proposed in this paper. SCA method has a self-aggregation property that the data objects will move towards each other to form clusters through SCA theoretically,which can reveal the inherent pattern of similarity hidden in the dataset. The inputs of SCA are just the pairwise similarities of the dataset,which makes it easier for time series analysis due to the variable length of the time series. Our experimental results on many problems have verified the effectiveness of SCA on some engineering application.