Purpose <p>Detrended fluctuation analysis (DFA) is an important technique for analyzing EEG time series and uncovering hidden trends and patterns. One of the key steps in DFA is the partitioning of the signal into nonoverlapping windows of the same size. The selection of the optimal window size is crucial because it affects the estimate of the scaling parameter, which describes the extent of correlations. However, the prior setting of the window size range in the DFA algorithm, which varies from user to user, can be a drawback that affects the accuracy and reliability of the analysis because it raises the possibility of subjectivity, especially if the scale setting is not sufficiently supported or confirmed.</p> Methods <p>To avoid this limitation, in this paper, we propose an improved version of DFA called automatic detrended fluctuation analysis (ADFA), which generates its own range of windows according to the length of the input signal, which makes it user-independent. To explore the efficiency of the presented framework, we first verified the ADFA with a set of synthetic signals and then tested it on real EEG data to separate between seizure and seizure-free intervals.</p> Results <p>We found that, compared with the DFA method, the ADFA method improved the classification accuracy by 10.75%.</p>