【摘要】 支持向量机的输出结果可以通过几何分析划分为六个连续的区间,并求得各个区间内训练样本的错误分类频率.本文以二分判别为例,将每个区间上的误分频率与logistic回归对预测样本的输出概率进行比较,提出了一种集成logistic回归与支持向量机的判别分析规则,并采用支持向量机效果验证的基准数据集进行实证分析.实证结果验证了所提出方法的有效性.更多还原

【Abstract】 By geometrical analysis,the value domain of the signed distance of support vector machines can be partitioned into six continuous intervals,and the frequency of misclassification in each interval could be calculated.Based on binary discriminant,we propose an integrated discriminant analysis rule(IDAR) through comparing the frequency of misclassification in each interval with the output probability of logistic regression.The validity of IDAR is illustrated by numerical results on several benchmark datasets.更多还原