Detection of Epileptic Seizures Using an Accelerometer Device & Machine Learning

Abstract

Epilepsy is the most prevalent neurological disesase in the world total which causes seizures in the patient. Detection of these seizures requires specialized approaches such as video/electroencephalography monitoring which are mainly available at specialized hospitals and institutes. Hence there is a need of developing a suitable system which can be made available to each patient and can accurately detect epileptic seizures. A wireless remote monitoring system based on a wrist-worn accelerometer is an optimum choice for the same. Mobile device based accelerometer sensors have been used to capture data and are capable of detecting seizures with short duration. Captured data is preprocessed using activity and time filtering to reduce the volume of data. Time domain features were extracted from the preprocessed data. Support vector machines, Logistic Regression and Random Forest Classifier were then used to classify non-seizure and seizure events. One-Class outlier detection algorithms such as One-class SVM and Isolation Forest also used to detect seizures as outliers as these were the minority data points in the data set A wearable accelerometer-based seizure detection system would aid in continuous assessment of convulsive seizures in a timely and non-invasive manner.