Multi-view Classification via Twin Projection Vector Machine with Application to EEG-Based Driving Fatigue Detection

• Published in: Data Mining and Big Data pp. 394 - 407
• Main Authors: Chen, Xiaobo, Gao, Yuxiang
• Format: Book Chapter
• Language: English
• Published: Singapore Springer Nature Singapore
• Series: Communications in Computer and Information Science
• Subjects: Driving fatigue detection; Multi-hyperplane model; Multi-view classification; Support vector machine
• ISBN: 9789811992964; 9811992967
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-981-19-9297-1_28

Multi-view learning based on a variety of multiple hyperplane classification (MHC) models has shown promising performance for multi-view data classification in recent years. However, seeking for a single fitting hyperplane for each class might be insufficiently expressive for the datasets with complex feature distribution. Moreover, in the presence of outlier data, most approaches tend to produce degraded results due to the adverse impact of outliers. In this paper, we put forward a new multi-view MHC model termed as multi-view twin projection vector machine (MvTPVM) which aims to seek for multiple projection vectors. Following the consensus principle, multi-view co-regularization is introduced to constrain the projected features of two views. To further achieve robust multi-view classification, we propose a robust variant called RMvTPVM where the distance involved in this model is measured by L1,2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L_{1,2}$$\end{document}-norm. To solve the resulting model, an elegant iteration algorithm is further proposed. The experimental results on both standard UCI datasets and driving fatigue detection based on EEG signals verify the effectiveness of our models in multi-view classification.