基于F-score的大数据公共空间模式选择方法

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.

F-score based common spatial pattern selection approachfor big data processing

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.