K-Means clustering based time series weighting with epileptic seizure detection

In detecting epilepsy and classifying epileptic attacks, the electrical activity of the brain is used as an important data source. In this study, the data set for automatic classification of epilepsy from EEG signals was taken from the UCI machine learning data repository. This dataset consists of raw time domain EEG signals. Apart from epilepsy, there are four different classes. These classes consist of a total of five classes, EEG recorded when the eyes are open, EEG recorded when the eyes are closed, EEG recorded from people with the tumor zone, and EEG recorded from healthy people. To differentiate the epileptic condition from the other classes, only the raw EEG signals were categorized without any feature extraction from the time domain EEG signals. In this study, k-averages cluster-based time series weighting (KOKTZSA) method was applied to raw EEG signals as pre­processing to classify the five-class epilepsy data set with high accuracy and then used to classify the weighted data set in Random Forest and C4.5 decision tree classification algorithms have been used. The raw EEG signal obtained a classification accuracy of 70.55% for the C4.5 decision tree classification algorithm and 96.86% for the data cluster C4.5 decision tree weighted by the KOKTZSA. Raw EEG signal, random forest classification algorithm obtained 81% classification accuracy while data set weighted with KOKTZSA achieved 99.33% classification accuracy with random forest classification algorithm. The obtained results show that the proposed hybrid model achieves a high classification accuracy without extracting any features from the EEG signal. This has greatly reduced the high computational cost.