Detecting and Visualizing Profile Correlation in Subspace

In this paper, we propose a novel method for detecting and visualizing profile correlation in subspace. The proposed method is able to (1) detect shifting-and-scaling correlated profiles in subspace, where the correlation can be either positive or negative, (2) summarize and visualize the shifting-and-scaling correlation in subspace, and (3) allow users to explore interested correlation subspace interactively. Initially, shifting and scaling were regarded as two different correlation patterns, in which profiles in subspace can overlap by a single shifting and scaling respectively. In later work, a much more generous correlation, shifting-and-scaling correlation, was studied, compared to which, shifting correlation and scaling correlation are just two special cases. Shifting-and-scaling correlation ensures subspace profile coherence not only in tendency as those tendency-based methods do, but also in value change proportion in subspace. However, no work has been focused on visualization of subspace shifting-and-scaling correlation yet. Our work is the first one to enable interactive exploration and visualization of subspace shifting-and-scaling correlation.