Transfer Learning for Subsequence Clustering from Multiple Source Domains

In many data-sparse applications, acquiring sufficient data for clustering can be challenging. To address the clustering of sparse subsequences within a target domain, this paper proposes a novel transfer learning subsequence clustering algorithm leveraging multiple source domains. Based on an improved Fuzzy C-Means algorithm, the method transfers knowledge from multiple source domains to guide target domain subsequence clustering. Crucially, canonical correlation coefficients between the target domain and each source domain are computed to determine the weight of transferred knowledge, thereby mitigating negative transfer. The proposed algorithm is evaluated extensively on diverse synthetic and real-world datasets. Experimental results demonstrate its superior performance compared to related clustering algorithms across multiple established public datasets and two real-world case studies.