Epileptic Seizure Prediction in EEG Records using Parallel Tree Based Learning and Feature Extraction

Abstract

Objectives: To propose an effective classifier to identify interictal from preictal signals and to reduce the EEG data size to fit into the processing model. Methods/Analysis: The first phase of the processing model identifies several features related to time-series data and identifies the features for data of specific dimension, called the epoch. A set of features are generated for each epoch and the corresponding class details are appended. The obtained data is used by the Classifier for the learning phase, followed by the test phase to identify the effectiveness of the classifier model. Findings: Experiments were conducted on EEG data obtained from American Epilepsy Foundation. The proposed features were extracted and the classifier model is trained and the prediction levels were recorded. It was observed that the true prediction levels of interictal signals showed high accuracy (0.91), while the true prediction levels of preictal signals exhibited moderate accuracy (0.66). The false predictions were identified to range from low to moderate (0.08 and 0.33). The accuracy levels were observed to be 0.75 and F-Measure was found to be 0.77. The architecture also exhibits moderate recall (0.66) and high precision levels (0.93). Novelty/Improvement: The proposed technique enables effective reduction of data by extracting features for an epoch, hence enabling customized data fine-tuning and scalability.