文献类型：会议论文

英文题名：A New Hybrid Feature Selection Algorithm Applied to Driver's Status Detection

作者：Ye, Peng-fei[1];Chen, Lan-lan[1];Zhang, Ao[1]

机构：[1]East China Univ Sci & Technol, Key Lab Adv Control & Optimizat Chem Proc, Minist Educ, 130 Meilong Rd, Shanghai 200237, Peoples R China

会议论文集：24th International Conference on Neural Information Processing (ICONIP)

会议日期：NOV 14-18, 2017

会议地点：Guangzhou, PEOPLES R CHINA

语种：英文

外文关键词：Driver's status detection; Multiple physiological signals analysis; Hybrid feature selection; Tabu search

摘要：This research introduces a framework based on multimodal feature analysis and hybrid feature selection algorithm for improving the recognition rate of driver's status. In order to provide rich information about physiological conditions of human operators, a variety of physiological features are widely extracted from time, spectral, wavelet and nonlinear domains. The redundant and noisy parts of the original feature set could negatively influence the identification performance and occupy limited computing resource. Therefore, a new hybrid feature selection approach is proposed to handle the high dimensionality of feature space and improve classification precision simultaneously. Decision Tree and Sparse Bayesian Learning were employed to generate the initial feature subset that could be further optimized by the adaptive tabu search with Fisher classifier. Finally, three-level driver's stress statuses were discriminated by using support vector machine. Our experimental results show that the proposed algorithm has achieved satisfactory identification rate of driver's status with compact feature vector.