Abstract: Too many feature attributes and the optimization of internal parameters are important factors that affect the generalization ability of Support Vector Machine(SVM)models.For the above two problems,to improve the generalization ability and classification accuracy of the SVM model,Random Forest(RF)algorithm and Quantum Particle Swarm Optimization algorithm(QPSO)were combined to optimize the SVM model.The importance of each feature was calculated with RF algorithm,and the features with high importance were selected as a feature subset through feature selection.Using the global search capability of QPSO algorithm,the optimal kernel function parameter  σ and penalty factor C of SVM model were found.The optimal parameters were applied to SVM model for classification prediction.Experimental simulation results showed that compared with other machine learning algorithms,the proposed model had higher training accuracy and prediction accuracy,which also confirmed that the model had better robustness and generalization performance.

Key words: random forest, quantum particle swarm optimization, support vector machine, feature selection, robustness

摘要： 特征属性过多及内部参数的优选是影响支持向量机(SVM)模型泛化能力的重要因素,针对以上两个问题,为了提高SVM模型的泛化能力和分类精度,将随机森林(RF)算法和量子粒子群优化(QPSO)算法结合优化SVM模型的核函数参数和惩罚因子。首先利用RF算法计算出每个特征的重要性,通过特征选择筛选出重要性高的特征作为特征子集;再利用QPSO算法的全局搜索能力寻找出SVM模型的最优核函数参数和惩罚因子,最后将最优参数应用于SVM模型中进行分类预测。实验仿真结果表明,与其他机器学习算法相比,所提模型具有较高的训练精度和预测精度,也证实了该模型具有较好的鲁棒性和泛化性能。

关键词: 随机森林, 量子粒子群优化, 支持向量机, 特征选择, 鲁棒性