TS3FCM: trusted safe semi-supervised fuzzy clustering method for data partition with high confidence

• Published in: Multimedia tools and applications 2022-04, Vol.81 (9), p.12567-12598
• Main Authors: Huan, Phung The, Thong, Pham Huy, Tuan, Tran Manh, Hop, Dang Trong, Thai, Vu Duc, Minh, Nguyen Hai, Giang, Nguyen Long, Son, Le Hoang
• Format: Article
• Language: English
• Subjects: Algorithms; Clustering; Computational efficiency; Computer Communication Networks; Computer Science; Computing time; Data points; Data Structures and Information Theory; Datasets; Machine learning; Multimedia Information Systems; Outliers (statistics); Performance degradation; Soft computing; Special Purpose and Application-Based Systems
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-022-12133-6

Data partition with high confidence is one of the main concentration of researchers in Soft Computing for many years. It is known that there may be some data with less confidence (wrong values, incorrect attribute types, irrelevant domain ranges, etc.) existed in the whole dataset due to the data gathering process. This would degrade the performance of final clustering results because of noises and outliers being occurred. Safe semi-supervised fuzzy clustering has been used extensively in recent years to tackle with this problem by adding the concept of a local graph between labeled and unlabeled data so that wrong labeled data has small impact to the final clusters. However, this process often takes much computational time and sometimes produces unreasonable results. In this research, we propose a new algorithm for the Data partition with confidence problem named as Trusted Safe Semi-Supervised Fuzzy Clustering Method (TS3FCM). The key motivation behind TS3FCM is to handle the drawbacks of the related safe semi-supervised fuzzy clustering algorithms regarding huge computational time. The novelty of TS3FCM against the other safe semi-supervised fuzzy clustering algorithms lies at the isolated processes of finding trusted labeled data and performing semi-supervised fuzzy clustering. The key contributions of the paper are briefly summarized as follows. At first, a new objective function is proposed. This function is incorporated with new weights for each labeled data so that the system can check whether a labeled data point is corrected or not. This function is also optimized to find the cluster centers and the membership matrix. Indeed, the labeled data having small impact after clustering are either set up with very low membership values or removed from the set of labeled data. Furthermore, a new semi-supervised fuzzy clustering model is defined to partition the whole dataset with the additional information being a mixture of the prior membership degrees ( U ¯ ) and labeled data. The whole TS3FCM works through 3 main phases with the main aim to accelerate the computational time and to achieve reasonable clustering quality compared to the related algorithms. TS3FCM is implemented and experimentally compared against the related methods such as the standard Fuzzy C-Means (FCM), the Semi-supervised Fuzzy Clustering method (SSFCM), and the Confidence-weighted safe semi-supervised clustering (CS3FCM) algorithm by both the computational time and the quality of cl