Automatic Graph Building Approach for Spectral Clustering

摘要: Spectral clustering techniques have shown their capability to identify the data relationships using graph analysis, achieving better accuracy than traditional algorithms as k-means. Here, we propose a methodology build automatically representation over input for spectral based approaches by taking into account local and global sample structure. Regarding this, both Euclidean geodesic distances are used main between given point neighboring samples around it. Then, information about structure, estimate an affinity matrix means of Gaussian kernel. Synthetic real-world datasets tested. Attained results show how our approach outperforms, in most cases, benchmark methods.