基于可信度阈值优化的案例推理评价分类方法

为了提高案例推理(CBR) 分类器的性能, 提出一种基于可信度阈值优化的CBR 评价分类方法. 首先, 通过一种可降低时间复杂度的改进型可信度评价策略对案例重用得到的建议解的可信度进行计算; 然后, 通过遗传算法(GA) 对可信度阈值进行迭代寻优; 接着, 根据得到的优化阈值将目标案例及其建议解划分为可信集或不可信集; 最后, 对不可信集按多数重用原则进行分类结论的调整, 从而实现可信的CBR 评价分类. 对比实验表明, 改进的可信度评价策略能有效提高分类性能, 从而可提高CBR分类器的决策与学习能力.

Trustworthiness evaluation method with threshold optimization for case-based reasoning classification

To improve the performance of a case-based reasoning (CBR) classifier, a trustworthiness evaluation method with threshold optimization for case-based reasoning classification is proposed. Firstly, an improved trustworthiness evaluation(TE) strategy is adopted to calculate the trustworthiness value of the suggested solutions achieved in reuse step. Then, the optimal threshold value of the trustworthiness is obtained by using the genetic algorithm(GA). Subsequently, the target case and its suggested solution is divided into the trustworthy set and the untrustworthy set in accordance with this threshold value. Finally, the majority reuse strategy is adopted to adjust the suggested solutions in the untrustworthy set so as to fulfill an overall CBR evaluation classification process. The experimental results show that the proposed method can effectively increase the classification performance and improve the learning ability for a CBR classifier.