基于加权支持向量机的多分类概率估计

Multiclass Probability Estimation via Weighted Support Vector Machine

多分类问题与概率估计在各大领域尤其是生物学和医学方面具有许多重要的应用。支持向量机作为在许多分类问题中都能取得高准确率的算法,却不能对类别进行概率估计。文章提出一种基于加权支持向量机的多分类概率估计算法,该算法将支持向量机作为基础模型,以分类准确率衡量算法的表现,利用不平衡数据对分类结果的影响,对分类样本的损失函数进行加权处理,根据权重求得类别的概率估计值。通过数值模拟和实证研究,验证了本文所提出的多分类概率估计方法在多分类问题中的分类预测效果显著优于其他通过概率估计进行分类的方法。

Multiclass classification problems and probability estimation have many important applications in various fields,especially in biology and medicine.As a highly accurate algorithm in many classification problems,support vector machines(SVM)cannot estimate the probability of classification.This paper proposes a multi-classification probability estimation algorithm based on weighted SVM.This algorithm uses the SVM as the basic model to measure the performance of the algorithm with classification accuracy,and employs the impact of the unbalanced data on the classification result to weight the loss function of classified samples,obtaining the probability estimate of the category according to the weight.Through numerical simulation and empirical study,the paper verifies that the proposed method of multi-classification probability estimation is significantly better than other methods of classification based on probability estimation.