Compositional Correlation for Detecting Real Associations Among Time Series

TL;DR

This paper introduces a new compositional correlation method that effectively detects both linear and nonlinear relationships in time series data by analyzing all possible data segment compositions, outperforming traditional correlation measures.

Contribution

The paper presents a novel compositional correlation approach that considers all data segment compositions, enhancing detection of complex relationships in time series data.

Findings

Successfully detects linear and nonlinear relationships.

Outperforms Pearson, Spearman, and distance correlation methods.

Identifies associations missed by traditional methods.

Abstract

Correlation remains to be one of the most widely used statistical tools for assessing the strength of relationships between data series. This paper presents a novel compositional correlation method for detecting linear and nonlinear relationships by considering the averages of all parts of all possible compositions of the data series instead of considering the averages of the whole series. The approach enables cumulative contribution of all local associations to the resulting correlation value. The method is applied on two different datasets: a set of four simple nonlinear polynomial functions and the expression time series data of 4381 budding yeast (saccharomyces cerevisiae) genes. The obtained results show that the introduced compositional correlation method is capable of determining real direct and inverse linear, nonlinear and monotonic relationships. Comparisons with Pearson's correlation, Spearman's correlation, distance correlation and the simulated annealing genetic algorithm maximal information coefficient (SGMIC) have shown that the presented method is capable of detecting important associations which were not detected by the compared methods.