粒子群和遗传算法优化神经网络结构的破产预测  

Bankruptcy Prediction Using Neural Networks with Structure Optimization by Particle Swarm Optimization and Genetic Algorithm

作　　者：杨钟瑾[1] 

机构地区：[1]广东财经大学信息学院,广东广州510320

出　　处：《数学的实践与认识》2015年第23期192-201,共10页Mathematics in Practice and Theory

摘　　要：引荐了运用粒子群算法和遗传算法优化多层前馈神经网络结构预测破产的方法.融合了粒子群算法、遗传算法和神经网络众多优点,自适应和并行地搜寻神经网络最优的结构,由此构建优化的预测模型.采用源自UCI机器学习数据库的破产和非破产混合样本数据集,随机地从数据集中读取数据并进行数据预处理,运用7重交叉校验方法客观地评价预测结果.仿真证明,方法能自动有效地构建神经网络的优化结构,具有更快的学习速度和更好的推广性能.与其它方法相比,方法具有更高的破产预测准确率.A method using particle swarm optimization algorithm and genetic algorithm to optimize structures of multilayer feedforward neural networks for bankruptcy prediction is presented. The presented method combines the advantages of particle swarm optimization, genetic algorithm and neural network, which adaptively searches for the optimal structure of neural networks in parallel way to construct optimal prediction model. A sample dataset comprised of bankruptcy and non-bankruptcy data derived from the UCI machine learning repository is used, the data are randomly read from the dataset and automatically prepro- cessed. A 7-fold cross-validation test is used to objectively evaluate the prediction results. The experimental results show that the presented method can automatically select the optimal structure of neural networks with good generalization capability and fast learning speed. A comparison of the prediction results clearly shows that the presented method has higher bankruptcy prediction accuracy than the other methods.

关 键 词：粒子群算法 遗传算法 神经网络 优化 结构 破产预测 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]