Index Financial Time Series Based on Zigzag-Perceptually Important Points

Abstract

Problem statement: Financial time series were usually large in size, unstructured and of high dimensionality. Since, the illustration of financial time series shape was typically characterized by a few number of important points. These important points moved in zigzag directions which could form technical patterns. However, these important points exhibited in different resolutions and difficult to determine. Approach: In this study, we proposed novel methods of financial time series indexing by considering their zigzag movement. The methods consist of two major algorithms: first, the identification of important points, namely the Zigzag-Perceptually Important Points (ZIPs) identification method and next, the indexing method namely Zigzag based M-ary Tree (ZM-Tree) to structure and organize the important points. Results: The errors of the tree building and retrieving compared to the original time series increased when the important points increased. The dimensionality reduction using ZM-Tree based on tree pruning and number of retrieved points techniques performed better when the number of important points increased. Conclusion: Our proposed techniques illustrated mostly acceptable performance in tree operations and dimensionality reduction comparing to existing similar technique like Specialize Binary Tree (SB-Tree).