Author
Abstract
Detrended Fluctuation Analysis (DFA) estimates the scaling exponent α of long-range correlated time series, but its polynomial detrending step fails when signals contain non-polynomial trends such as exponential drifts, regime changes, or periodic oscillations. In this paper, we propose Neural-DFA, a modification of the DFA algorithm that replaces polynomial detrending with a trained neural network while keeping the rest of the DFA framework intact. The neural detrender is a 1D U-Net conditioned on observation scale via Feature-wise Linear Modulation (FiLM), which learns a detrending function that recovers α more accurately under non-polynomial trend contamination than polynomial detrending of any tested order. We train the network in two phases: first on synthetic segments where the true trend is known, and then through a differentiable DFA pipeline that allows gradients to flow from the estimated α back into the network weights, directly optimizing the accuracy of the scaling exponent. On a composite nonlinear trend benchmark using fractional Gaussian noise (fGn), Neural-DFA achieves a mean squared error of 0.0014, which is 35% lower than DFA-4 and 15 times lower than DFA-3. At low Hurst exponents where trend contamination dominates, the improvement over DFA-4 reaches a factor of 15. We validate the method on real-world data from three domains: heart rate variability (HRV) recordings from PhysioNet, financial volatility series, and global temperature anomaly records. Neural-DFA shows strong agreement with DFA-4 across all signals (Pearson r=0.980) while producing estimates closer to values established in literature on HRV data and showing systematic divergence from polynomial DFA on signals with suspected non-polynomial contamination. These results demonstrate that Neural-DFA is a practical alternative to polynomial detrending for fractal analysis of signals with complex nonlinear contamination.
Suggested Citation
Kubin, William, 2026.
"Neural-DFA: Learned nonlinear detrending for Detrended Fluctuation Analysis,"
Physica A: Statistical Mechanics and its Applications, Elsevier, vol. 698(C).
Handle:
RePEc:eee:phsmap:v:698:y:2026:i:c:s0378437126005145
DOI: 10.1016/j.physa.2026.131778
Download full text from publisher
As the access to this document is restricted, you may want to
for a different version of it.
Corrections
All material on this site has been provided by the respective publishers and authors. You can help correct errors and omissions. When requesting a correction, please mention this item's handle: RePEc:eee:phsmap:v:698:y:2026:i:c:s0378437126005145. See general information about how to correct material in RePEc.
If you have authored this item and are not yet registered with RePEc, we encourage you to do it here. This allows to link your profile to this item. It also allows you to accept potential citations to this item that we are uncertain about.
We have no bibliographic references for this item. You can help adding them by using this form .
If you know of missing items citing this one, you can help us creating those links by adding the relevant references in the same way as above, for each refering item. If you are a registered author of this item, you may also want to check the "citations" tab in your RePEc Author Service profile, as there may be some citations waiting for confirmation.
For technical questions regarding this item, or to correct its authors, title, abstract, bibliographic or download information, contact: Catherine Liu (email available below). General contact details of provider: http://www.journals.elsevier.com/physica-a-statistical-mechpplications/ .
Please note that corrections may take a couple of weeks to filter through
the various RePEc services.