Structure Identification-Based Clustering According to Density Consistency

Algorithm 1

The process of new clustering algorithm.

Input: Data set 𝑋, in which each row is a pattern; parameters: 𝑘, ⧵alpha, ⧵beta.

Output: Label vector 𝐿, in which each element is cluster label.

  Step  (1): Find 𝑘 nearest neighbors for each data point 𝑥 according to

      Euclidean distance.

  Step  (2): Find nearest neighbors for each data point 𝑥 according to

      Definition 3.2.

  Step  (3): Construct filtering matrix 𝑈 based on (3.2) and (3.8).

  Step  (4): Filtering process 𝑋=𝑈𝑋, obtain elementary clusters, which are

      the element with various kind of structure features. Top-down

      process is included to identify connectedness and direction.

  Step  (5): Integrate elementary clusters with same structure features into

      meaningful clusters according to (3.9).

  Step  (6): Top-down process to identify noisy data points and outliers.