论文部分内容阅读
针对传统学习矢量量化算法没有考虑属性的重要度差异的问题,提出一种加权学习矢量量化算法.该算法为每一维属性引入一个权重系数,用其表征相应属性在分类过程中的重要程度,并与权向量一同更新.利用输入样本和获胜神经元之间的修正距离的均值,控制权重系数更新的阈值及步长.距离均值确保了更新过程的稳定性,且无需进行权重系数的归一化操作.UCI机器学习数据库中6组数据的实验结果表明,该算法能够有效给出数据的本质属性,尤其是局部型权重系数.与传统学习矢量量化算法及其改进算法相比,识别率高、性能稳定、计算复杂度低.
Aiming at the problem that traditional learning vector quantization algorithm does not consider the importance difference of attributes, this paper proposes a weighted learning vector quantization algorithm, which introduces a weight coefficient for each dimension attribute, and uses it to represent the importance of the corresponding attributes in the classification process, And updated together with the weight vector.Using the average value of the modified distance between the input sample and the winning neuron, the threshold and the step length of the weight coefficient update are controlled.Distance mean ensures the stability of the updating process without the need to normalize the weighting coefficient The experimental results of 6 sets of data in the machine learning database of UCI show that this algorithm can effectively give the essential attributes of the data, especially the local type weight coefficient.Compared with the traditional learning vector quantization algorithm and its improved algorithm, the recognition rate is high , Stable performance, low computational complexity.