A Sparse Binary Data Clustering Method for Transaction Data

Cluster analysis is an important approach in data mining, yet binary data has received less attention compared to categorical and numerical data. Thus, this study intends to address this gap by concentrating on the clustering of sparse binary data with high dimensions. It introduces a novel method that integrates the sparse binary dimensionality reduction method with a particle swarm optimization (PSO) algorithm-based fuzzy K-modes algorithm. The proposed approach not only achieves dimensionality reduction but also explores the optimal cluster centroids through the application of the PSO algorithm. Furthermore, both within-cluster variance and maximum entropy are combined as the objective during the clustering process. To evaluate the performance of the proposed method, a case study involving transaction data clustering is undertaken. The empirical findings demonstrate the superiority of the proposed method over existing ones when applied to sparse binary data, as evidenced by improved fitness values.