A Hybrid Algorithm for Clustering of Time Series Data Based on Affinity Search Technique

Pseudocode 1

Pseudocode related to TTC.

Method: (𝐷, 𝛼, 𝐾)

Input:  𝐷: the set of time-series 𝐷=𝐹₁,𝐹₂,…,𝐹_𝑛

    𝛼: Affinity threshold

    𝐾: cluster number

Output: 𝐶: set of clusters

/*𝑆𝑡𝑒𝑝  1. time-series data reduction*/

(1)  𝐷 =𝑧-norm(𝐷) /*𝑧-normalization of all time-series */

(2)  𝐴[𝑁][𝑁]←Smilarity_ED (𝐷) /* Calculate disimilarity array based on (5)*/

(3)  (SC[1to𝑀],  𝑎[1to𝑁]) ← CAST(𝐴, 𝛼)  /*  𝑀 is determined automatically by CAST*/

(4)  for  𝑖=1 to 𝑀; 𝑀 is the number of sub-clusters

(5)    𝑟 ← Average (SC_𝑖, 𝑎[1to𝑁]) /* summarize the cluster */

(6)    𝑅 ← 𝑅 ∪ 𝑟/*𝑅: a collection of prototypes  */

(7)  end for

/*𝑆𝑡𝑒𝑝2. clustering*/

(8)  𝐵 [𝑀][𝑀]←Smilarity_DTW(𝑅)

(9)  𝐶^′← 𝑘-medoids(𝑘,𝐵)

(10) 𝐶← labels(𝐶^′)validate the MTC clustering results

(11) return 𝐶