摘要
以扩张矩阵理论为基础 ,应用数学规划理论提出了一种规划模型求解方法 ,可以更好地实现概念学习和特征提取。与传统的启发式算法相比 ,采用遗传算法求解的规划模型可以找到多个全局最优解以及可行解。实例计算表明了该方法的有效性。
Learning from examples and feature subset selection are the two basic problems and also the bottle neck in concepts extraction of machine learning. Based on extension matrix formed on positive and negative examples, we set up the integer programming model (IPM) for optimal rule extraction and feature subset selection. The IPM method can find the multiple optimal solution in theory and practice, and experiments reveal that IPM , while solved by genetic algorithms, runs more efficiently compared with heuristic algorithms as FCV, GS, HCV, GFS, etc.. We have used IPM as the core method of knowledge acquisition in data mining and knowledge discovery in databases about financial analysis and tax auditing, and got effective results.
出处
《系统工程学报》
CSCD
2000年第2期163-167,207,共6页
Journal of Systems Engineering
基金
国家自然科学基金!资助项目 ( 7940 0 0 13 )
关键词
示例学习
扩张矩阵
知识获取
机器学习
特征选择
learning from examples
extension matrix
inter programming model
knowledge acquisition
machines learning
