Scalable Distributed Laplacian Score for Feature Selection

Abstract

Feature selection has been extensively explored in supervised learning contexts. Conversely, it is much more challenging in unsupervised learning due to the lack of class labels that facilitate the identification of relevant information. The Laplacian Score is an important metric for feature selection that can work efficiently without the need for class labels. However, traditional computation of the Laplacian Score is computationally intensive and not feasible for large-scale datasets. In this paper, we presented a scalable approach for computing the Laplacian Score for feature selection. The proposed method significantly improves the time complexity of the original Laplacian Score algorithm, making it feasible for large-scale datasets. The proposed scaling mechanism uses Apache Spark, which enables the Laplacian Score to be computed in a timely and memory-efficient manner. The proposed algorithm retains the original accuracy and interpretability of the Laplacian Score. Experimental results demonstrate substantial time improvements, highlighting the efficiency of the proposed approach in practice. Our contributions provide a crucial step towards broader adoption of the Laplacian Score in real-world applications, where computational efficiency and scalability are essential. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.