Investigation on AQ11, ID3 and the Principle of Discernibility Matrix

• 作 者: 王珏 崔佳 赵凯
• 出 处: Journal of Computer Science and Technology
• 发表日期: 2001年01期
• 关键词: discernibility matrix knowledge representation machine learning decision tree

The principle of discernibility matrix serves as a tool to discuss and analyze two algorithms of traditional inductive machine learning, AQ11 and ID3. The results are: (1) AQ11 and its family can be completely specified by the principle of discernibility matrix; (2) ID3 can be partly, but not naturally, specified by the principle of discernibility matrix; and (3) The principle of discernibility matrix is employed to analyze Cendrowska sample set, and it shows the weaknesses of knowledge representation style of decision tree in theory.

The principle of discernibility matrix serves as a tool to discuss and analyze two algorithms of traditional inductive machine learning, AQ11 and ID3. The results are: (1) AQ11 and its family can be completely specified by the principle of discernibility matrix; (2) ID3 can be partly, but not naturally, specified by the principle of discernibility matrix; and (3) The principle of discernibility matrix is ​​employed to analyze Cendrowska sample set, and it shows the weaknesses of knowledge representation style of decision tree in theory.