基于改进天牛群算法优化SVM的个人信用评估  

Personal Credit Evaluation Based on SVM Optimized by Improved Beetle Swarm Optimization Algorithm

作　　者：陈静静 刘升[1] CHEN Jing-jing;LIU Sheng(School of Management,Shanghai University of Engineering Science,Shanghai 201620,China)

机构地区：[1]上海工程技术大学管理学院,上海201620

出　　处：《计算机技术与发展》2021年第6期135-139,共5页Computer Technology and Development

基　　金：上海工程技术大学硕士研究生科研创新项目(19KY0321)。

摘　　要：由于支持向量机(SVM)的分类性能受参数影响较大,为了提高SVM在个人信用评估中的精度,提出基于改进天牛群算法优化SVM的个人信用评估方法。在对天牛的速度更新时加入天牛的自身判断,更加贴合生物觅食本性;通过改进收缩因子来优化学习因子,更好地协调局部与全局搜索之间的平衡;引入正态分布函数,自适应调整步长,改善算法收敛速度慢且易陷入局部极值的缺点。利用获取的较优参数来构建分类模型,进而提高SVM的分类性能。采用UCI中的4个数据集,并与其他参数优化方法进行对比,实验证明IBSO-SVM具有较高的寻优性能。为了验证改进模型在信用评估方面的性能,首先通过随机森林对信用数据German的特征进行了筛选,随后对处理过的数据进行实例分析,结果证明了混合模型的有效性。The classification performance of support vector machine(SVM) is greatly affected by parameters. In order to improve the accuracy of SVM in the personal credit evaluation, a personal credit evaluation method based on SVM optimized by improved beetle swarm algorithm is proposed. When updating the speed of the beetle, the self-judgment of the beetle is added, which is more consistent with the nature of biological foraging. By improving the contraction factor to optimize the learning factor, the balance between local search and global search can be better coordinated. The normal distribution function is introduced to adjust the step, so as to amend its deficiencies, such as slow convergence speed and easy falling into local optimum. The optimal parameters are used to construct the classification model to improve the classification performance of SVM. The experiment uses four UCI data sets and compares them with other parameter optimization methods. It is showed that IBSO-SVM has a high optimization performance. Finally, the verification is based on the German dataset which is processed by random forest. The experimental results prove the effectiveness of the improved algorithm.

关 键 词：支持向量机 信用风险评估 天牛群算法 参数优化 随机森林 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]