基于F-score的大数据公共空间模式选择方法  

F-score based common spatial pattern selection approach for big data processing

作　　者：王欣杰[1] 李海峰[1] 马琳[1] 吴明权[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,黑龙江哈尔滨150001

出　　处：《燕山大学学报》2014年第5期432-439,共8页Journal of Yanshan University

基　　金：国家自然科学基金资助项目(61171186;61271345);语言语音教育部-微软重点实验室开放基金资助项目(HIT.KLOF.20110XX);中央高校基本科研业务费专项资金资助项目(HIT.NSRIF.2012047)

摘　　要：公共空间模式(CSP)分析由于具有变换简单、实现快速等优点,被广泛应用于信息挖掘、脑电信号处理等大数据处理应用中。本文以基于脑电信号的人类认知状态分类识别为背景,研究一种基于Fisher分数(Fscore)的特征评价与选择的CSP构建方法。利用F-score计算代价小、可以快速从高维数据中选择出有效信息和特征的优点,实现了对模式重要程度做出定量的表达;针对F-score阈值确定困难、信息冗余、无法自适应实现等难点问题,提出了相应的解决方法。所提出方法在脑认知活动解析实验中,针对五类认知状态分类问题取得了92%的识别准确率。本方法为大数据的公共模式挖掘等问题供了一个强有力的新工具。Due to such advantages like simplicity and high speed, the common spatial pattern （CSP） analysis method has beenwidely applied in various big data processing applications such as information mining, brain signal processing, etc. Facing thehuman cognitive state recognition problem, a F-score based hybrid feature evaluation and selection method is investigated for CSPdiscovery and construction. Since F-score may easily and quickly to pick up the effective features fromhigh dimensional data, theproposed F-score based method quantitatively represents the importance of each data pattern. Other solutions are also proposed toconquer the conventional F-score problems like threshold definition difficulty, Information redundancy, lack of automatic and selfadaptive,etc. In the cognitive state analysis experiments, the proposed method obtained a recognition accuracy of 92%on 5 kindsof cognitive tasks, proving it to be a nouvelle and powerful tool for public pattern mining from big data.

关 键 词：公共空间模式 信息选择 F-score 认知状态识别 大数据 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]