Entropy Variability Factor Test

The proposed Entropy Variability Factor (EVF) framework can reliably detect subtle structural changes in the local irregularity of time-series data. Specifically, we hypothesize that:
Rolling Shannon entropy can serve as a sensitive local measure of disorder.
Combining this with a two-sample Z-test and an empirically derived ±2% null zone provides a robust statistical test to distinguish genuine changes from random background variation.
The EVF framework performs consistently across synthetic and real-world datasets and remains stable under a range of practical parameter settings.This shared dataincludes all MATLAB scripts and input data files used to implement the proposed Entropy Variability Factor (EVF) framework as presented in the manuscript.
The data includes:
Synthetic signals: These are generated under controlled conditions to simulate known structural patterns, constant variance scenarios (stable conditions), and abrupt variance shifts (change detection scenarios).
Real-world data: Includes a time series from publicly available COVID-19 community mobility datasets (Google Community Mobility Reports), segmented around the onset of COVID-19 to test the framework on real behavioral changes.
Contents:
1. Benchmark 1: MATLAB script demonstrating the intuitive response of rolling Shannon entropy for simple signals with known structures.

2. Benchmark 2: MATLAB script for testing the EVF framework under constant-variance conditions to establish the empirical ±2% null zone.

3. Benchmark 3: MATLAB script for detecting controlled synthetic variance shifts, showing the test’s detection performance.

4. Benchmark 4: MATLAB script for applying the EVF framework to real-world COVID-19 mobility data, including signal segmentation and test output.
5. Benchmark 5: MATLAB script for applying the EVF framework to real-world 2021 Texas power crisis -Feb-2021 data, including signal segmentation and test output.
6. Sensitivity Analysis: Scripts that systematically vary the number of histogram bins, rolling window sizes, and segment lengths to assess the framework’s robustness across realistic parameter ranges
Notes:
1. All codes were developed and run using MATLAB R2022a.
2. Some scripts and functions may require additional time to execute, especially when processing large time-series with high-resolution rolling windows and multiple bin configurations.
3. The provided code and data have been tested for reproducibility and transparency. Users are encouraged to adjust the window size, bin count, and other parameters as needed to replicate the benchmarks or adapt the Entropy Variability Factor (EVF) test for their own datasets at different variables.