基于AQ覆盖框架的蚁群规则集学习算法

Ant colony rule set learning algorithm based on AQ covering frame

针对规则集学习问题,提出一种遵循典型AQ覆盖算法框架(AQ Covering Algorithm)的蚁群规则集学习算法(Ant-AQ)。在Ant-AQ算法中,AQ覆盖框架中的柱状搜索特化过程被蚁群搜索特化过程替代,从某种程度上减少了陷入局优的情况。在对照测试中,Ant-AQ算法分别和已有的经典规则集学习算法(CN2、AQ-15)以及R.S.Parpinelli等提出的另一种基于蚁群优化的规则学习算法Ant-Miner在若干典型规则学习问题数据集上进行了比较。实验结果表明:首先,Ant-AQ算法在总体性能比较上要优于经典规则学习算法,其次,Ant-AQ算法在预测准确度这样关键的评价指标上优于Ant-Miner算法。

A novel ant colony rule set learning algorithm（Ant-AQ） is presented based on the combination of AQ covering frame and ant colony optimization.The ant colony optimization substitutes for the beam search in the specification procedure of AQ coveting frame.This strategy can reduce occurrence of convergence to solutions coding local optima for evaluating Ant-AQ,the algorithm is applied to several typical rule set learning problems and compared to the classical algorithms for rule set learning （CN2,AQ-15） and another rule set learning algorithms based on ACO called Ant-Miner which proposed by R.S.Parpinelli et.al. The experiment results show,first,the algorithm has much better overall performance than classical algorithms mentioned above, and second, the algorithm has advantages, over Ant-Miner on the key criteria of prediction accuracy.