應用感知特徵點法於財務時間序列的群集分析

Clustering Financial Time Series Based on Perceptually Important Point Calculation

趙子安(Tsu-An Chao)

指導教授 ： 羅淑娟

國立臺北科技大學/管理學院/資訊與運籌管理研究所/碩士(2013年)

https://doi.org/10.6841/NTUT.2013.00213

摘要

時間序列資料普遍出現於生活中的各個領域，學術研究上也不乏許多學者利用資料探勘技術針對時間序列資料進行群集分析，因為時間序列通常隨著時間成長且資料維度大，很難直接發現各筆資料之間的關聯性，不利於資料分析，許多研究企圖利用群集分析將類別未知的資料進行分群，從中發現各個群集內的特性並說明各筆資料之間的關聯性，如此將簡化對時間序列資料分析的複雜度。本研究結合感知特徵點擷取法(Perceptually Important Points)與動態時間校正(Dynamic Time Warping)以及階層式群集法(Hierarchical Clustering)建構一金融市場輔助選股決策的程序，協助投資人能夠更精確地掌握相似度高的投資標的，本研究期望股市時間序列資料經過PIP將雜訊降低之後能提昇序列資料進行相似度比對時的凝聚性。
實驗資料顯示，DTW校正雜訊的程度相較於PIP法是比較不足，透過PIP特徵減量之後減少了圖形的複雜程度以及雜訊，提昇了資料凝聚的效果，藉由實證市場資料台灣五十，我們可以得知PIP特徵減量後降低資料雜訊的能力，以及利用凝聚法得到不同階層的凝聚結果，透過本研究程序確實可掌握股價走勢相似度高的個股。

關鍵字

時間序列 ； 動態時間校正 ； 感知特徵點 ； 階層式群集法

並列摘要

It is usual to observe time series data appearing in many fields such as science, engineering, business, finance, economic and health care. Many researchers use data mining skills to cluster time series data in their studies. Time series data easily become huge and highly dimensional with time cumulated. It is not easy to analyze time series data directly, due to the complexity of time series data. Therefore, in this study we attempted to reduce the dimension of time series before clustering the data. Our study used the Perceptually Important Point (PIP), Dynamic Time Warping (DTW), and Hierarchical Clustering to construct a systematic procedures to search the stocks with same trend for investors. The results showed that PIPs have better similarity and clarity than those without noise reduction. Investors can get more detail levels of the clustering results from hierarchical clustering chart if needed. Based on our procedure, investors can clearly and easily observe those stocks with the similar trends under different clustering levels.

並列關鍵字

Time Series ； Dynamic Time Warping ； Perceptually Important Point ； Hierarchical Clustering

參考文獻

[3]林俊良、郭雅玲，應用馬可夫決策過程探討當機機台最佳之修復策略，科技管理學刊，Vol. 10，No.2，pp. 31-56，2005。

[4]黃柏傑，以群集式基因演算法為基礎的時間序列切割法與型樣之發掘， 碩士，國立成功大學，台南市，2008。

[5]B. Zhou, J. Hu ,"A dynamic pattern recognition approach based on neural network for stock time-series," Nature & Biologically Inspired Computing, 2009, pp.1552-1555.

[6]C. Guo, H. Li, D. Pan ,"An improved piecewise aggregate approximation based on statistical features for time series mining," Lecture Notes in Computer Science (Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), 2010, pp.234-244.

[8]F. L. Chung T. C. Fu, V. Ng, R. W. P. Luk, "An evolutionary approach to pattern-based time series segmentation," IEEE Transactions on Evolutionary Computation, Vol.8, no.5, 2004, pp.471-489.