Two-mode probabilistic distance clustering

Caner, Yağmur
Probabilistic Distance Clustering (PDC) is a soft clustering technique constructed around some axioms. It is a center-based approach and assigns each data point to multiple clusters with a membership probability. The PDC is applicable for one-mode data sets, where each data points’ quantitative or qualitative values over each feature are stored. This study focuses on PDC and consists of two main contributions. Firstly, the relevance of PDC to some other probabilistic models in the literature is examined. We show that PDC method and its axioms explain models from marketing, location theory, and unsupervised learning. Secondly, this thesis proposes two original solution methods for the soft Two-Mode Clustering (TMC) problem. Two-mode clustering is a technique to cluster two-mode data, representing a linkage between two sets of data points. A comprehensive computational study is conducted on continuous, noisy, and binary data sets. The use of membership probabilities for decision-making is also discussed. This study will be the pioneer soft assignment approach for two-mode clustering literature.
Citation Formats
Y. Caner, “Two-mode probabilistic distance clustering,” M.S. - Master of Science, Middle East Technical University, 2021.