Combining Semi-supervision and Hubness to Enhance High-dimensional Data Clustering

Abstract

The curse of dimensionality turns the high-dimensional data analysis a challenging task for data clustering techniques. Recent works have efficiently employed an aspect inherent to high-dimensional data in the proposal of clustering approaches guided by hubs which provide information about the distribution of the data instances among the K-nearest neighbors. Though, hubs can not well reflect the implicit data semantics, leading to an unsuitable data partition. In order to cope with both issues (i.e., high-dimensional data and meaningful clusters), this paper presents a clustering approach that explores the combination of two strategies: semi-supervision and density estimation based on hubness scores.
The experimental results conducted with 23 real datasets show that the proposed approach has a superior performance when applied on datasets with different characteristics.