Tri-level robust clustering ensemble (TRCE) algorithm for clustering enhancement with comparative analysis

This paper introduces a Tri-Level Robust Clustering Ensemble (TRCE) algorithm that leverages the Three-Staged Clustering Algorithm (TSCA) as a fundamental component in its advancement. The purpose of TRCE is to enhance the clustering outcomes' quality by initially generating multiple clustering outcomes at various θ values. Subsequently, a voting mechanism is employed to consolidate the outcomes produced by each θ value. The fundamental aim of such a process is to ensure that the generated clustering results have maximized the ratio of agreements between the samples within each generated cluster. The TRCE has been verified on some benchmark datasets and then compared with several individual algorithms such as Kmeans, K-prototype, and squeezer algorithms. It has also been compared with ensemble-based algorithms such as selecting initial seeds based on the co-association matrix (SICM), selecting initial seeds based on previous results (SIPR), Average Normalized Mutual Information (K-ANMI), Clustering Categorical Dataset (CDC) and Clustering Categorical Data by Cluster Ensemble (ccdByEnsemble). Some supervised clustering algorithms were also involved in the comparison such as ICKDC, SKDEKMean, and ISSKDEKMeans. The experimental results showed the strengths of the TRCE in terms of the clustering quality over the compared clustering algorithms. It was always ranked highly and managed to handle numerical, categorical, and mixed-type datasets.