基于核独立分量分析的模糊核聚类神经网络集成方法*

提出一种基于核独立分量分析的模糊核聚类神经网络集成方法。该方法首先采用核独立分量分析对高维数据进行特征提取；随后用模糊核聚类算法根据相互独立训练出的多个神经网络个体在验证集上的输出对其进行分类，并计算每一类中的所有个体在验证集上的泛化误差；然后取其中平均泛化误差最小的神经网络个体作为这一类的代表；最后经相对多数投票法得到集成的最终输出。实验结果表明，与其他集成方法相比，该方法具有较高的精确度和稳定性。

Neural network ensemble based on KICA and KFCM

This paper proposed neural network ensemble based on KICA and KFCM. First analyzed the data with high dimension patterns using KICA to extract main features. Then used the KFCM algorithm to classify the individual neural network training independent by its output on validate sample and calculated the generalization errors of neural networks in each category to the validate set, according to which the individual neural network whose generalization error was minimum in its category will be selected and combined the predictions of the component networks through majority voting. The experiments show that the proposed approach has higher accuracy and stability compared with other ensemble methods.