Incremental Learning System for Disease Prediction: Epileptic Seizure Case

摘要

Analysis of statistical characteristics in electroencephalogram (EEG) data can accurately predict epileptic seizures. However, traditional on-off line prediction methods ignore the specificity of different epileptic subjects (e.g., age and seizures region in brain) and the diversity of epileptic seizure modalities (e.g., seizures during stress and awake state). which leads to low prediction accuracy and poor flexibility. In this paper, sparse group penalty algorithm and incremental learning mechanism are proposed to improve seizure risk prediction. In particular, sparse group penalty algorithm is proposed based on data correlation to incorporate the dependence structure among the features into solving approaches. Then, a relative spectral feature extraction approach is applied to construct a pattern library incrementally. Further, the prediction model parameters are dynamically updated and adjusted based on the updated subject-specific pattern library and incremental learning mechanism. The experimental results show that the proposed epileptic seizure prediction mechanism can sparse the parameters of the model and reduce the retraining time of the parameters. At the same time, it has high prediction accuracy and robustness.