一种基于反例样本修剪支持向量机的事件追踪算法

支持向量机（SVM）在各类别样本数目分布不均匀时，样本数量越多其分类误差越小，而样本数量越少其分类误差越大．在分析这种倾向产生原因的基础上，提出了一种基于反例样本修剪支持向量机（NEP—SVM）的事件追踪算法．该算法首先修剪反例样本，根据距离和类标决定一反例样本的取舍，然后使用SVM对新的样本集进行训练以得到分类器，补偿了上述倾向性问题造成的不利影响．另外，由于后验概率对于提高事件追踪的性能至关重要，而传统的支持向量机不提供后验概率，本文通过一个sigmoid函数的参数训练将SVM的输出结果映射成概率．实验结果表明NEP—SVM是有效的．

Event Tracking Algorithm Based on Negative-Example-Pruning Support Vector Machine

When training sets with uneven class sizes are used, the larger the sample size, the smaller the classification error of support vector machine (SVM), whereas the smaller the sample size, the larger the classification error. A negative-examples-pruning support vector machine (NEP-SVM) based algorithm for event tracking was proposed based on the analysis of the cause of this bias. The algorithm first pruned the negative examples, reserved and deleted a negative sample according to distance and its class label, then trained the new set with SVM to obtain a classifier and this algorithm compensates for the unfavorable impact caused by this bias. In addition, since posteriori probability of samples was important in improving the performance of event tracking, but traditional SVM did not provide posteriori probability, so the parameters of a sigmoid function were trained to map the SVM outputs into probabilities in this paper. Experimental results showed that the NEP-SVM is effective.