论文部分内容阅读
针对传统二叉树在多分类问题上存在分类精度不够高和时间复杂度较高的不足,提出了一种基于二叉树结构双优化的SVM多分类学习算法。此算法利用遗传算法对已经提取的特征参数子集和核参数进行双重优化,以获得最优的主要特征参数,从而有效地解决了样本结构复杂、分布不平坦的多分类识别问题。作者运用UCI数据库中的数据,通过仿真实验,并就经度和时间复杂度与有向无环图法和一对一法作比较,结果表明本文提出的算法具有较好的优越性。
In order to overcome the shortcomings of traditional binary tree in multi-class classification, such as high classification accuracy and high time complexity, a multi-class SVM learning algorithm based on binary tree structure is proposed. This algorithm uses genetic algorithm to double optimize the extracted subset of feature parameters and kernel parameters to obtain the optimal main feature parameters, so as to effectively solve the multi-classification identification problem with complex sample structure and uneven distribution. The author uses the data from UCI database, and through simulation experiments, compares the longitude and time complexity with directed acyclic graph method and one-to-one method. The results show that the proposed algorithm has better advantages.