基于多元数据图表示的广义统计模式识别

Generalized Statistical Pattern Recognition Based on ultivariate Visualization

现代统计模式识别方法以数据满足一定的统计分布规律为前提,然而现实问题研究中存在大量分布不理想或小样本情况。基于多元图表示的广义统计模式识别提出基于类别样本统计分布特性分析来选择狭义统计或非统计方法,它主张统计学方法在模式识别领域的科学运用。首先介绍了广义统计模式识别概念,然后基于人工构造数据集对该方法进行了数据仿真实验,结果显示,类别样本统计分布特性差异对分类方法选择具有显著影响。

Statistical pattern recognition techniques are exercising dominion over nowadays. But there are many real-world problems the sample is not obey the known Statistical models. Generalized statistical pattern recognition method selects proper technique based on the distribution estimation test of samples. It suggests scientific applying the statistical method to pattern recognition and avoiding the distortion and aberration of information because of the statistical method abuse. Firstly, some concerned concepts of special nonstatistic were introduced. Then gave the process of generalized statistical pattern recognition. Manual creating the experiment data based on finite normal mixture model, the experiment results show that the error rate of special non-stastical classifier is lower than special statistical classifier for the sample of finite normal mixture model, and there are no distinct difference of normal distribution sample.