Detrended Fluctuation Analysis (DFA) provides insights on signal complexity which have shown to be relevant and effective for dis tinguishing healthy and non-healthy persons through different physiological signals. This method is based on different steps involving linear regression. This paper proposes an evidential linear regression model and its application on DFA in order to take into account limitations of DFA due to uncertainties associated with crisp linear regression estimates. Three R packages that contains a DFA implementation has been compared with the proposed method in experiments realised on sinusoidal signals, noises and Hausdorff famous dataset that illustrates the interest of the method.