A1192
Title: Fuzzy clustering with L0 penalization for mixed-type data
Authors: Maria Brigida Ferraro - Sapienza University of Rome (Italy) [presenting]
Marco Forti - University of Rome "Sapieza" (Italy)
Paolo Giordani - Sapienza University of Rome (Italy)
Andrea Nigri - University of Foggia (Italy)
Abstract: A novel penalized fuzzy clustering method for mixed-type data is proposed. Building on the Fuzzy $K$-Prototypes ($FKP$) framework, it incorporates an $L_0$-penalized membership matrix, promoting sparsity by selectively reducing active cluster assignments. This refinement improves interpretability and clustering efficiency by minimizing noise and enforcing more distinct group structures. To handle mixed-type data, Euclidean distance is applied to numerical variables, while categorical variables are clustered using a similarity-based measure. The inclusion of $L_0$-penalization provides a powerful tool for analyzing heterogeneous datasets. Its applications span diverse fields - including economics, social sciences, and biomedical research - where uncovering meaningful patterns is crucial.
