Automated Tonic-Clonic Seizure Detection Using Random Forests and Spectral Analysis on Electroencephalography Data

<p>Artificial intelligence (AI) has a potential for impact in the diagnosis of neurologi-</p> <p>cal conditions, the academic consensus generally has a positive outlook regarding how AI can</p> <p>improve the care of stroke victims and those who suffer from neuro-degenerative conditions such</p> <p>as dementia. When combined with Internet of Things technology, this could facilitate a new par-</p> <p>adigm for epilepsy treatment. These technologies have applications in improving the welfare of</p> <p>epileptics, epilepsy being a common neurological condition that can result in premature death</p> <p>without a quick response. As such it is important for the system to avoid false negatives. This</p> <p>investigation focused on how machine learning algorithms can be utilised to identify these events</p> <p>through Electroencephalography (EEG) data. The UCI/Bonn dataset, a classic benchmark for</p> <p>automated epilepsy detection systems was identified and utilised. This investigation focused on</p> <p>the random forest algorithm. Given that EEG neurological data represents time series data and</p> <p>machine learning excels at this task, automation could be achievable via a wearable device. From</p> <p>there, Fast Fourier Transforms (FFT) were applied to identify if spectral features of EEG signals</p> <p>would aid identification of seizures. This method achieved an accuracy of 99%, precision of 98%</p> <p>and a recall of 100% in 12.2 milliseconds time to classify and one second of EEG data. These</p> <p>results show that random forests combined with FFT are a viable technique for attaining high</p> <p>recall when detecting grand mal epileptic seizures in short periods of time. CHB-MIT dataset</p> <p>was utilized for parity also showing good performance.</p>