【摘要】 本文将模式识别中的特征选择问题转化为有向图上最佳路径搜索问题,并应用AI中的Best First(简记BF)策略搜索最佳路径,提出了特征选择GBFF和TBFF算法,证明了用它们可不穷举而一定找到最佳子集,同目前被认为最好的全局最佳算法——B&B相比,TBFF搜索的特征子集数目优于B&B.更多还原

【Abstract】 In this paper, the problem of feature selection is converted into the optimal pathsearching problem in a weighted directional graph. Then by means of the so called informed Best First (BF) search strategy for problem solving in AI., Algo(?)ithms GBFF and TBFF are proposed to search the optimal path, i.e., the optimal feature subset. These algorithms guarantee optimality of the selected subset without exhaustive search. In compararison with the well known Branch and Bound(B & B) algorithm, it b(?) been shown that the number of the expanded modes by TBFF is less (even much less) than that by B & B; In other words, TBFF is superior to B & B.更多还原