基于标准序列浮动前向特征选择的改进算法研究

Research on Improved Algorithm Based on The Sequential Floating Forward Selection

摘要:

信息技术的高速发展促进了信息领域内涵的根本性变革,信息特征的表述方法和内涵不断扩充,高维特征大幅涌现;这些高维特征中可能存在许多不相关和冗余特征,造成了维度灾难,这对基于特征空间聚散特性的分类识别算法提出了更高的要求,需要利用特征选择算法,降低特征向量维数并消除数据噪音的干扰;针对高维特征向量引入的维度灾难等问题,围绕目标分类识别的具体应用,基于标准的序列浮动前向特征选择算法,完成交叉验证重复次数优化,提出了改进的特征选择算法;通过仿真实验表明,基于Bayesian分类器开展识别时,改进算法能够在确保分类识别正确率的前提下,有效提升特征选择的计算速度,并维持一个相对更为收敛且稳定的置信区间,具备良好的准确度。

Abstract:

With the rapid development of information technology, the indicative method on the information characteristics keep expanding,high-dimensional feature emerge and grow with a massive trend. These high-dimensional feature contain much redundant and irrelevant feature, which will result in the curse of dimensionality. This situation will further lead to higher requirements and more challenges for the classification and recognition algorithm,need the feature selection algorithm to reduce the dimension of eigenvector and data noise.Aim at the dimension disaster introduced by the high dimension eigenvector,and the application oriented ATR algorithm,porpose an improved algorithm based on the sequential floating forward selection,by optimizing the repeat number of cross-test.The results of the simulation experiments shows that on the premise of the high classification accuracy,this improved algorithm can upgrade the calculation speed effectively and could maintain a more astringent and more stable confidence interval what means a better accuracy.