基于敏感度分析的案例特征项权重算法的改进  

Improved Algorithm for Case Feature Weight Based on Sensitivity Analysis

作　　者：杨宝华[1] 辜丽川[1,2] 李绍稳[1] 

机构地区：[1]安徽农业大学信息与计算机学院,合肥230036 [2]合肥工业大学管理学院,合肥230009

出　　处：《计算机科学》2010年第5期194-196,209,共4页Computer Science

基　　金：国家自然科学基金(30800663);国家863高科技计划(2006AA10Z249);安徽省高校省级自然科学研究项目(KJ2007B158;KJ2008B111)资助

摘　　要：研究案例库特征项权重的确定方法,通过集成BP神经网络和敏感度分析,改进案例库特征项的权重确定算法,将案例库中的各特征项和决策目标项构造一个BP神经网络,经训练和学习后,依次删除输入节点,分析网络的输出对输入的敏感程度,确定各特征项的权重。并以红籽西瓜仁重的案例库对其进行测试,结果表明该算法是有效的。Method for weight of case-based reasoning feature was studied. A method based on I3P neural network and sensitivity analysis was put forward, and the algorithm of the weight for the feature in case was improved. BP neural network about feature and decision was constructed, after training and learning, the input nodes were deleted one by one, the output of sensitivity was analysed and weight was calculated. The case on Red-seed Watermelon was tested,which shows the algorithm is effective.

关 键 词：敏感系数 特征项 权重 神经网络 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]