Local Correlation Visual Analysis Based on Subspace Clustering

The dimension correlations which exist in subset of data are often obscured in the full dataset. We propose a local correlation visual analysis approach to detect meaningful data subset and reveal local dimension correlations. First, a scatter plot is adopted to visually reveal the subspace cluster. The two dimensions of the scatter plot are defined based on geodesic distance and the distance between local subspaces correspondingly. Next, an estimation for correlation significance is proposed based on covering area and mean distance of the data. Subsequently, the 2-dimensional subspaces which reveal local correlations are suggested. Last, a visual analysis system is implemented and case studies demonstrates the effectiveness and efficiency of our system.