“…In particular, the execution times for files of durations between 15 and 30 minutes are between 8.8 seconds and 131.7 seconds, that is execution times between 6.8 to 204.5 times lower than the duration of the original EEGs. Furthermore, the method doesn't impose any condition on the signals to be compared as it handles the cases where EEGs to be compared 7 autoregressive model 8 We use an AR model instead of an ARIMA model since it is argued in [101] that "for every ARIMA model there exists an equivalent AR model, that can be obtained from the ARIMA model by polynomial division" and the LPC cepstrum of the time series is, subsequently (in [101] ), computed based on that equivalent AR model 2.12063087537 )) using the Levenberg-Marquardt algorithm Moreover, fractal interpolation doesn't require model selection as AR modeling does, which considerably speeds up EEG interpolation. Moreover, with our dataset, the computation of the Euclidean distance between the cepstrum coefficients calculated based on the EEGs AR models leads to a matrix of NaN (i.e not a number) 9 : the AR modeling method is therefore less stable than the fractal interpolation-based method.…”