预测型关联规则演化学习的适应值函数

Fitness Function for Evolutionary Learning of Predictive Association Rules

为了提高基于遗传算法的分类预测准确度,探讨了评价规则质量的适应值函数,提出了基于置信度和支持度加权和的适应值函数,以取代传统的基于灵敏性和选择性的适应值函数.理论分析和实验结果都表明,文中提出的新适应值函数对于预测型关联规则演化搜索的引导作用明显地优于传统的适应值函数.新的适应值函数有利于改进基于遗传算法的机器学习.

In order to improve the predicted accuracy of the classification based on genetic algorithm, the fitness functions for evaluating the rule equality are discussed in this paper. A fitness function based on a weighted sum of confidence and support is then proposed, which can substitutes the traditional fitness function based on sensitivity and specificity. Both the theoretically analytical and the experimental results show that the proposed fitness function has a greater advantage in the evolutionary search of predictive association rules over the traditional one, and is helpful to the improvement of the machine learning based on genetic algorithm.