Online Differential Evolution for Sparse Feature Selection in Data Streams

Implementations of high-dimensional streaming data often employ online streaming feature selection (OSFS) techniques. However, practical implementation frequently encounters issues of incomplete data caused by equipment failures and technical limitations. Online Sparse Streaming Feature Selection (OS²FS) addresses this challenge by employing missing data imputation through latent factor analysis. Current OS²FS methods demonstrate significant limitations in feature evaluation, leading to performance degradation. To overcome these limitations, this paper proposes a novel online differential evolution for sparse feature selection (ODESFS) in data streams, which incorporates two key innovations: 1) imputing missing values by leveraging a latent factor analysis (LFA) model; 2) applying differential evolution (DE) to evaluate feature importance. Comprehensive experiment on six real-world datasets demonstrates that ODESFS outperforms pioneering OSFS and OS²FS methods, consistently achieving higher accuracy by selecting optimal feature subsets.