Adaptive Initialization Method Based on Spatial Local Information for 𝑘-Means Algorithm

Algorithm 1

AIF-SLI-𝑡-NN.

Input: The dataset 𝒳.

Output: Candidate set 𝐶_cand.

(1) Initial candidate set 𝐶_cand=∅;

(2) Construct the Gram matrix 𝑊_𝑖𝑗 based on (2);

(3) Compute vector 𝑑^→ with 𝑑_𝑖=∑^𝑛_𝑗=1𝑊_𝑖𝑗;

(4) Determine a threshold 𝜔=median(𝑑^→);

(5) for all data points in 𝒳

(6)  if data point 𝑥_𝑗 with local density 𝑑_𝑗>𝜔,

(7)   Mark data point 𝑥_𝑗 as a candidate initial center, that is 𝐶_cand=𝐶_cand∪𝑥_𝑗;

(8)  end if

(9) end for