分支界定算法及其在特征选择中的应用研究  

Study of Branch&Bound Algorithm and Application of Feature Selection

作　　者：王思臣[1] 于潞[1] 刘水[1] 唐金元[1] 

机构地区：[1]海军航空工程学院青岛分院,山东青岛266041

出　　处：《现代电子技术》2008年第10期142-144,共3页Modern Electronics Technique

摘　　要：分支界定算法是目前为止惟一既能保证全局最优,又能避免穷尽搜索的算法。他自上而下进行搜索,同时具有回溯功能,可使所有可能的特征组合都被考虑到。对分支界定算法进行研究,并对其做了一些改进;最后对改进前后的算法在特征选择领域进行比较,选择效率有了明显的提高。Branch＆Bound Algorithm is the only method which can ensure best of all the vectors, and it can avoid endless searching. It searches from top to bottom and has the function that from bottom to top, so it can include all of the feature vectors. The Branch＆Bound Algorithm is studied in the paper,and it is improved, the two algorithms are compared by the feature seclection,the efficiency of seclection is improved greatly.

关 键 词：分支界定算法 特征选择 特征集 最小决策树 局部预测 

分 类 号：TP31[自动化与计算机技术—计算机软件与理论]