基于LVQ神经网络的改进覆盖算法

Improved covering algorithm based on LVQ neural network

覆盖算法是一种具有高分类准确度和强泛化能力的构造性神经网络分类算法。针对其选择覆盖中心的随意性,结合竞争性神经网络方法对覆盖算法进行改进,在覆盖学习之前进行预学习,选择最佳覆盖球形中心,来优化覆盖。通过标准UCI测试数据实验的比较,从分类的准确性和覆盖个数方面进行对比,得到改进的覆盖算法有很好的效果。

Covering algorithm is a classification algorithm with constructive neural network of high accuracy and strong generalization ability. According to the arbitrariness of the selections of cover center, combining competitive neural network method, an improved algorithm of covering algorithm is proposed. Before learning covering algo- rithm, learning center samples of the sample set as cover centers are selected by LVQ to optimize covering spheri- cal. In the experiment in the test data standard UCI, comparing two parameters： classification accuracy and the cov- ering number, the result of experiments shows that the proposed method is effective.