Cross-domain, soft-partition clustering with diversity measure and knowledge reference

Area of Science:

• Machine Learning
• Data Mining
• Clustering Algorithms

Background:

• Conventional soft-partition clustering methods like FCM, MEC, and FC-QR struggle with insufficient or noisy data.
• Existing models often lack robustness and effective handling of outliers.

Purpose of the Study:

• To propose novel fuzzy clustering models, QWGSD-FC, TI-KT-CM, and TII-KT-CM, that address limitations of conventional methods.
• To enhance clustering performance in low-data or high-noise environments using transfer learning.
• To ensure privacy preservation in cross-domain clustering tasks.

Main Methods:

• Development of the quadratic weights and Gini-Simpson diversity based fuzzy clustering (QWGSD-FC) model.
• Introduction of two transfer learning-based frameworks: Type-I (TI-KT-CM) and Type-II (TII-KT-CM) knowledge-transfer-oriented c-means.
• Utilizing historical cluster centroids and fuzzy memberships for knowledge transfer.

Main Results:

• QWGSD-FC effectively handles intra-cluster deviation and provides unbiased probability assignments.
• TI-KT-CM and TII-KT-CM demonstrate high clustering effectiveness and parameter robustness in target domains.
• TII-KT-CM shows superior cross-domain performance due to comprehensive knowledge learning from source domains.
• Both TI-KT-CM and TII-KT-CM offer strong privacy protection by not requiring raw source data.

Conclusions:

• The proposed QWGSD-FC, TI-KT-CM, and TII-KT-CM models significantly improve upon conventional fuzzy clustering techniques.
• These novel methods are effective for clustering insufficient or noisy data and in cross-domain scenarios.
• The developed algorithms offer robust performance, enhanced privacy, and superior clustering accuracy.