EEG-MINE: Mining and Understanding Epilepsy Data

Given electroencephalogram time series data from patients with epilepsy, can we find patterns and regularities? The typical approach thus far is to use tensors or dynamical systems. Here, we present EEG-MINE, a nonlinear, chaos-based "gray box model", that blends domain knowledge with data observations. When applied to numerous, real EEG sequences, EEG-MINE (a) can successfully reconstruct the signals with high accuracy; (b) can spot surprising patterns within seizure EEG signals; and (c) may provide early warning of epileptic seizures.