About
Multi-view datasets contain multiple forms of data and are beneficial for prediction models. However, using such data can lead to high-dimensional challenges and poor generalization. Selecting relevant features from multi-view datasets is important for addressing these issues and improving interpretability. Traditional feature selection methods have limitations, including failure to leverage information from multiple modalities and lack of generalizability. They are often tailored to specific tasks and lack interpretability.
To overcome these limitations, a novel approach called the multi-view multi-objective feature selection genetic algorithm (MMFS-GA) is proposed. It simultaneously selects optimal feature subsets within and between modalities, demonstrating superior performance and interpretability for multi-view datasets. Evaluation results on benchmark datasets show improvement over baseline methods. This work provides a promising solution for multi-view feature selection and encourages further research in this area.
Creators
Vandad Imani
