Title
Trend-aligned feature correlation: holistic feature relevance metric on time series data
Authors
Runhai He,1 Qingqing Song,1* Quanhua Zhou,1 Zhenxing Zhang,1 Shoujing Zhang1
1 Belarusian State University (Belarus)
Published In
Second International Conference on Big Data, Computational Intelligence, and Applications (BDCIA 2024)
Published Date
20 March 2025
Abstract
Existing time series correlation research methods often rely too much on overall statistical characteristics and ignore the dynamic changes of data in the time dimension, especially in the identification of trend turning points and complex correlations. This study proposes an innovative trend-aligned feature correlation matching method (TAFC), which breaks through the limitations of traditional correlation analysis. TAFC adopts a sophisticated data preprocessing process and introduces two unique metrics: trend overlap index metric (Q) and numerical correlation metric (P). The P metric effectively captures the numerical correlation between data by combining the normalized difference measurement (NDM) and the Pearson correlation coefficient. In addition, TAFC also introduces a custom granular smoothing mechanism to balance trend feature preservation and noise suppression, aiming to comprehensively evaluate the multi-dimensional and multi-granular correlation of time series data. This study uses simulated scenario data based on chi-square distribution and multimodal distribution for experimental verification, and the verification range includes 36 sets of simulation data. The results show that this method can significantly distinguish data sets with different trend characteristics. The introduction of granularity has been verified to help make trend features more explicit. At the same time, when 5% random noise is introduced, the algorithm error is controlled within 5.60%-12.24%. This study provides a novel and accurate method for computer-aided time series data analysis, which has important application value.
Identifiers
- DOI: 10.1117/12.3059782
- EI-Compendex: 20251318141293
- Scopus: 2-s2.0-105001128637
Links
SPIE Digital Library
https://www.spiedigitallibrary.org/conference-proceedings-of-spie/13550/1355007/Trend-aligned-feature-correlation--holistic-feature-relevance-metric-on/10.1117/12.3059782.short
Harvard ADS
https://ui.adsabs.harvard.edu/abs/2025SPIE13550E..07H
Comment