Epileptic seizure detection using machine learning techniques

Abstract

EEG is commonly used to accommodate information about the electrical activity of the brain, an automated epilepsy seizure detection and prediction technique. As the rapid innovation in the field of healthcare increases, various biomedical signals namely, electrocardiogram (ECG), electroencephalogram (EEG), and electromyogram (EMG) play a crucial role in the measurement of various diseases such as cardiovascular diseases and brain disorders. Also, detection of epilepsy by an expert mainly neurologists, from EEG requires a lot of time as large amount of data is generated by the signals. We processed EEG database of CHB-MIT (scalp EEG), to find if our model could outperform the state-of-the-art models. The present study is mainly composed of four parts. In the first part, multiscale principal component analysis has been applied for EEG de-noising. In the second part, Discrete Wavelet Transform (DWT) is used to decompose EEG signal into a set of amplitude modulated and frequency modulated components. In the third part, features such as standard deviation, kurtosis, skewness, mean, power of coefficients, ratio of power of coefficients have been extracted from various mode functions and these six features have been categorized into three subsets. In the last stage, these subset of features are employed as inputs to 10 machine learning classifier for classification between non-seizure and seizure EEG signals. The simulation results show that our model achieved an accuracy of 70.30% for KNN, 95.47% for Random Forest, 92.56% for Decision Trees, 91.22% for Support Vector Machine, 84.12% for Multilayer Perceptron, 84.12% for Naive Bayes, and 88.92% for Ensemble Classifiers. Our model achieved maximum accuracy of 95.47% and sensitivity of 96.34% for Random Forest in ictal versus inter-ictal EEG for CHB-MIT database which is better when compared to existing state-of-the-art methods.