Detection of Epileptic Seizure Using Accelerometer’s Time Series Data and Hidden Markov Model

Abstract

As per a survey done worldwide, it is found that there are approx 50 million people living with epilepsy (0.5% to 1% of the average population), and the percentage are comparatively far more in developing countries to that of the developed countries. The count of India goes to approx 10 million which is 20% of the world epilepsy patients. Epileptic seizures happen due to brain dysfunction, which can be read by electrical activity in the brain as it assumed that there is sudden rush of electrical activity during seizures. When we visualise an automatic seizure detection system an accelerometer based device would be more practical because of its ease of wearability and usability. This project presents Hidden Markov Model probabilistic approach to detect the epileptic seizure. The accelerometer-based signal has been taken and features are extracted and then from the extracted feature relevant features are selected. Two different models for non-seizure and seizure data have been created. After the training of the module when any test data comes it passes through both the model and score is calculated depending upon the score classification of test data is decided. The result has also been compared with deep learning approach (LSTM) and it is found with the given peculiarities of data (highly unbalanced and small size) the performance of HMM is better than LSTM. It is expected that the project will become a vital tool for the implementation and research work of epileptic seizure detection.