DFCN-CE3: A Three-Way Clustering Algorithm Based on Deep-Fusion Clustering Network

As an unsupervised learning method, deep clustering has received widespread attention and shown remarkable results for processing high-dimensional samples in the fields of data mining. However, most of the existing deep clustering algorithms define one cluster by a set with sharp boundary, which fails to capture the uncertainty and ambiguity prevalent in real-world datasets. To address this problem, this paper introduces a contraction-and-expansion based deep fusion three-way clustering network (DFCN-CE3 for short). The proposed method uses deep fusion clustering network to obtain hard clustering result of the universe. Base on the hard clustering result, DFCN-CE3 shrinks each cluster to the core region through a contraction operation, and enlarges each cluster to the support set through an expansion operation. By the contraction–expansion strategy, a three-way clustering result is formed. Comparisons of experimental results with 8 clustering algorithms on 6 data sets validate the effectiveness of the proposed approach.