摘要
文章提出了一种组合聚类分析和神经网络的预测方法。聚类分析将大的数据集聚类划分为几类小的数据集,这样在每一类中,数据的相似度比较高,然后再分类训练相应的模型,最后做预测。建立加入聚类分析的径向基神经网络模型,用金融时间序列做试验,并跟径向基神经网络模型进行比较。试验结果表明,加入聚类分析的径向基神经网络模型提高了连续预测的趋势准确率,降低了时间代价,并减小了模型的复杂度。
A predicting method combining clustering and neuro-network is advanced and investigated. The big data group is divided into some small parts by clustering. By this way, every small part has a higher similarity degree, and we use data in these small parts to train corresponding model and do predicting. After constructing the RBF neuro-network added with the clustering method, we perform an experiment with it on the financial time series, and compare it with the primitive RBF neuro-network. The result of this experiment shows that the modified RBF neuro-network increases trend accuracy in sequential predicting, while debasing the cost of time and reducing the complexity of the model.
出处
《微电子学与计算机》
CSCD
北大核心
2006年第9期85-87,90,共4页
Microelectronics & Computer
关键词
聚类
时间序列
预测
径向基
神经网络
Clustering, Time series, Predicting, RBF, Neuro-network