anilsener / time-series-medicine

Code and documentation for seizure prediction from EEG data

time-series-medicine

The aim of this project is to study time-series data in healthcare and medical data.

Contents

Folder	Contents
Docs	Contains basic documentation about the project
EEG	Contains study of the Seizure Forecast problem
HAR	Contains study of the Human Activity Recognition (HAR) problem

The first dataset is EEG signals from subjects with epilepsy. The data is a collection of segments which contain recordings of signals in multiple channels. Each segment is to be classified as preictal (oncoming seizure) or interictal (no seizure). Two main approaches are adopted for the task of classifying the segments:

• Manually engineered features used to train a classifier
• Long short term memory (LSTM) networks which take the (time-averaged) signals as inputs

The data is available for downlaod from Kaggle. The data contains clips of EEG time-series data corresponding to interictal (no oncoming seizure) and preictal (oncoming seuizure) pieces. Each clip is provided in '.mat' format and needs to be converted into numpy arrays. This is achieved by the functions in utils.

The second dataset is the Human Activity Recognition dataset from the UCI repository.