Electroencephalography signal classification using neural network, decision tree and ensemble bagged tree for epilepsy disease

Abstract

Epilepsy is a brain disease caused by abnormal brain activities. Machine learning algorithms are usually applied in the classification and identification of epilepsy at an early stage. This study's primary objective is to classify the Electroencephalography (EEG) signal dataset of epileptic seizures using a machine learning algorithm and to evaluate the performance using the Plot Confusion Matrix and area under the receiver operating characteristic (AUC-ROC) curve. The plot confusion matrix method will give an array that depicts the True Positives, False Positives, False Negatives, and True Negatives. Besides, the AUC-ROC curve is a performance measurement for classification problems at various threshold settings. These methods can be used to check or visualize the performance of the multi-class classification problem. This thesis involves a collection of datasets containing 200 healthy individuals and 300 epilepsy patients. Next, features were extracted from these datasets. Feature extractions help to reduce data dimensionality and eliminate noise, while its output is used as the input for classifier-based epileptic classification. This study selected the Discrete Wavelet Transform (DWT) and Statistical Features as feature extraction methods. In addition, multiple machine learning techniques are presented in this study. Feed Forward Neural Network (FFNN), Back Propagation Neural Network (BPNN), Decision Tree, and Ensemble Bagged Tree (EBT) were used as classifiers. Furthermore, Linear Discriminant Analysis (LDA) has been selected as the benchmark standard. Therefore, five classifiers were trained for classification purposes. Each classifier is combined with DWT and Statistical Features. The proposed feature extraction, DWT combined with BPNN, gives the highest accuracy of 91.2% and a shorter duration of training.