Weighted Multi-view Feature Selection with Genetic Algorithm

Abstract

The feature selection problem of multi-view data has received widespread attention from researchers in recent years. However, existing multi-view approaches suffer from two main issues in weight optimization: (1) Weight coupling problem, where the weights of different views may be coupled, meaning that changing the weight of one view may affect the weights of other views. In such cases, the weight optimization process may be constrained, leading to suboptimal weight allocation. (2) Lack of handling instability, where some algorithms may not fully consider the instability in the weight optimization process, such as noise and changes in data distribution. This can result in unstable weight selections that cannot cope with uncertain data environments. To address these issues, we propose Weighted Multi-view Feature Selection with Genetic Algorithm (WMFS-GA). Specifically, our algorithm combines feature selection results from multiple views and encodes the selected features as initial features. This enables a more comprehensive utilization of information from multiple views, improving the accuracy and robustness of feature selection. We then employ an improved genetic algorithm for weight optimization, allowing for reasonable weight allocation for features from different views during the feature selection process, enhancing the integration and accuracy of multi-view data. Experimental comparisons with several state-of-the-art multi-view feature selection algorithms demonstrate significant advantages in classification performance for our proposed algorithm. Code for this paper available on: https://github.com/boredcui/WMFS-GA.