Figure 1.

Our methodological approach. The EEG data is first segmented into epochs (see A₁, A₂, A₃). Next, 58 features are extracted and an ensemble of unsupervised outlier detection methods are used (see B₁, B₂, B₃) to identify EEG epochs that are artifact-ridden and require interpolation (see A₂ and B₂). The artifact-ridden epochs are then interpolated by an ensemble of deep encoder-decoder networks (see red line in C).