摘要
现实中产生的时间序列数据中一般包含杂讯而且维度很高。符号化此类时间序列数据使得其维度降低,使得发现序列中的模式变得更简单。谱聚类是一种利用图的拉普拉斯矩阵特殊性质的聚类方法。相对于传统的k-means聚类,谱聚类在实际应用中有着更多优势。在那些展现出非凸性质的数据中,谱聚类一般会得到比k-means更符合常识的类。在股票数据使用上述方法,结果显示这种方法是有效的。
Real world time series data is notorious for its noise and high dimensionality. Symbolize time series greatly remedy this problem, and makes finding patterns easier. Spectral clustering is a family of similar algorithms that utilizes the special properties of graph Laplacian matrix. Compare with traditional k - means clustering, spectral clustering has many advantages in practice. On certain datasets where the data exhibit non - convex properties, spectral clustering often results in more sensible clusters than k - means clustering. In this paper, we experiment this idea on stock data and show the effectiveness of this approach.
出处
《九江职业技术学院学报》
2017年第2期14-16,共3页
Journal of Jiujiang Vocational and Technical College
关键词
机器学习
时间序列分析
聚类
machine learning, time series analysis, clustering
