Incremental unsupervised feature selection for dynamic incomplete multi-view data

Multi-view unsupervised feature selection has been proven to be efficient in reducing the dimensionality of multi-view unlabeled data with high dimensions. The previous methods assume that all views are complete. However, real applications, often incomplete, i.e., some instances missing, which will result failure these methods. Besides, while arrive form streams, existing suffer issues storage cost and expensive computation time. To address issues, we propose an Incremental Incomplete Unsupervised Feature Selection method (I2MUFS) on incomplete streaming data. By jointly considering consistent complementary information across different views, I2MUFS embeds into extended weighted non-negative matrix factorization model, can learn a consensus clustering indicator fuse latent matrices adaptive view weights. Furthermore, introduce incremental learning mechanisms develop alternative iterative algorithm, where is incrementally updated, rather than recomputing entire updated from scratch. A series experiments conducted verify effectiveness proposed by comparing several state-of-the-art experimental results demonstrate efficiency terms metrics computational cost.