学习样本存在分类错误时的决策判据分析

Analysis of Decision Rules in the Presence Of Classification Errors of the Learning Sample

在模式识别理论中,如果分类学习样本给定,随机观测序列的条件分布概率密度已知,那么该序列的分类问题可利用学习判据解决这一判据是在假设学习样本分类没有偏差的情况下,用极大似然估计参数替代贝叶斯定理中的真值进行的。讨论了学习样本的分类可靠性难以确信时的模式识别问题,研究了利用条件密度极大似然估计进行贝叶斯风险决策时,学习样本分类错误对它的影响,以及在此情况下的判据稳定性问题。

Pattern recognition problems are considered for the case of inaccessible credible classification of the learning sample. The effect of errors in classification of the learning sample on the risk of the Bayes decision rule is studied in the case of estimating the maximal likelihood of the conditional density parameters. Analytically and numerically solvable decision rules stable with errors in the learning sample are proposed and discussed.