基于特征筛选的云分类器

摘要

To improve the classification accuracy for multi-dimension independent samples, a feature selection method based on feature weight is presented. The weight of features is computed according to the function describing the similarities between features, and the features with less importance are eliminated according the computed weight, and so to solve the dimensionality reduction problem for multi-dimension samples. The selected features are consistent with those obtained using principle component analysis method. Finally the cloud classification model is constructed based on the selected features and is applied in the classification processes of iris database and complex ore images with good effects.%高维且不独立的样本特征集使分类的准确性降低,笔者提出一种根据样本集特征权值进行特征选择的方法.根据特征间的相似性度量函数计算特征的权重,并根据权重去除重要性差的特征,用于解决高维样本集的特征降维问题,特征选择结果与主成份分析结果一致.建立基于保留特征加权的云分类模型,应用于iris数据集和复杂矿石图像的分类,效果良好.