分解型分类数据挖掘算法研究及改进

Research and advances of splitting rule extraction data mining algorithm based on neural networks

神经网络由于具有良好的非线性映射能力和对任意函数的准确逼近能力,用于分类问题中的分类规则提取往往能获得很高的精度.本文针对一种分解型基于前馈网络的数据挖掘算法做了深入研究,给出了算法流程.根据其结构复杂的特点对前端输入做出了优化,并基于JOONE(Ja-va Object Oriented Neural Network)构造了RBF(Radial Basis Function)径向基分类神经网络,且通过UCI数据集验证了该方法的有效性.

For the advantages of neural networks, the capability of non-linear mapping and exact approaching make a higher precision of rule extraction exist in classifying problems. After figuring out the steps based on a kind of splitting rule extraction data mining algorithm, optimize the input nodes according to the complex structure. And build up a RBF classifying neural networks using JOONE, the UCI sample datum results shows that the effectiveness of this method.