无向马尔科夫毯分类器与集成

Undirected Markov blanket classifier and integration

无向马尔科夫毯结构是属性和类变量之间的最重要依赖结构之一,建立无向马尔科夫毯分类器的核心是无向马尔科夫毯结构学习。针对现有无向马尔科夫毯结构学习方法具有低效率和可靠性,以及不具实用性等问题,基于贝叶斯网络理论、马尔科夫网络理论和依赖分析方法进行具有多项式复杂度的无向马尔科夫毯结构和分类器学习,来避免这些问题。并建立最优性定理、可转换定理、可靠性定理和局部化定理为其提供理论依据。同时,对小例子集情况,给出了近似学习方法,并将无向马尔科夫毯分类器扩展为联合分类器,以有效地进行小例子集分类。

The undirected Markov blanket structure is one of the most important dependency structures be tween attribute and class variables. The key problem of learning undirected Markov blanket classifier is to build the undirected Markov blanket.structure. At present, the methods of learning undirected Markov blanket structure are of low efficiency and low reliability and are unpractical. The learning method of both undirected Markov blanket structure and classifier with polynomial complexity is presented based on the theory of Bayesian net- works, the theory of Markov networks and the dependency analysis way. The problems above can be avoided. The built optimal theorem, transferable theorem, reliability theorem and local theorem are the theory foundation of the presented method. The approximate learning arithmetic is developed and an undirected Markov blanket classifier is extended as a unite classifier to suit for small data classification.