A common application of non-parametric spectral density estimation is electroencephalogram (EEG) processing. The spectrum of an EEG signal carries a significant amount of information about the brain activity of a patient. Some of this can be found from the power of the frequency components in the signal. The frequency range of an EEG signal is often split into frequency bands or wave patterns. For instance high power in the gamma frequency band is thought to be linked to epileptic patients. By applying a spectral density estimation technique, information about a patient’s brain activity can be analyzed by looking at the power of the frequency components in each wave pattern. This information can then be used to get an idea of how the patient’s brain is behaving, and whether to diagnose them with epilepsy.
The aim of the project is to analyze epileptic activity using non-parametric density estimation methods on EEG signals. There is a publicly available EEG dataset online that contains five sets of EEG data, they are labeled as sets A, B, C, D, and E. Sets A and B contain EEG data from healthy patients with their eyes open (set A) and eyes closed (set B). Sets C, D and E are taken from patients who suffer from epilepsy. Sets C and D are taken during seizure-free intervals, while set E is taken during epileptic activity. Our aim is to apply different spectral density estimation techniques to analyze signals from the EEG dataset and compare their ability to extract useful information that could help predict which set the data came from (A, B, C, D or E).
Once the spectral density of an EEG signal has been found, it can be classified using an algorithm such as a k-nearest neighbours algorithm (KNN). KNN is a type of neural network that can be trained to extract relevant features from a dataset to classify it. The technique used is called supervised training, where the model is first trained by passing sets of data that the correct category is known and adjusting weights within the model to minimize the prediction error. Once the model has been trained, it can be tested on more datasets to assess its ability to correctly classify datasets it has not ‘seen’ before. The accuracy of the KNN classification will depend on the estimator's ability to extract useful information from the raw EEG data.
