Learning to learn decision trees

摘要: Decision trees are widely used in machine learning and knowledge acquisition systems. However, there is no optimal or even unanimously accepted strategy of obtaining "good" such trees, most the generated suffer from improprieties, i.e. inadequacies representing knowledge. The final goal research reported here to formulate a theory for decision domain, that set heuristics (on which majority experts will agree) describe good tree, as well specifying how obtain trees. In order achieve this we have designed recursive architecture system, monitors an interactive system based on driven by explanatory reasoning, incrementally acquires using it build domain theory. This also represented may be dependent. Our define notion good/bad measure their quality, needed guide constructing partial acquired at Each moment basic its tree generation process, thus constantly improving performance.