Time-Series Analysis in Epilepsy Prediction using Temporal Convolutional Networks (TCN)

Time-series analysis in epilepsy prediction involves analysing temporal patterns in neural activity such as EEG signals to detect changes that precede seizures. This work aims at evaluating the performance of Temporal Convolutional Networks (TCN) for time-series analysis of the EEG signal in epilepsy prediction. Because of the ability to model intricate temporal dependencies. Initially, are gathered from archives and clinical investigations and the input EEG signals are pre-processed to remove noise artefacts, scaled and segmented into frames. Feature extraction methods are applied to derive key characteristics from the signals, followed by Principal component analysis (PCA) and recursive feature removal are two feature selection techniques that are used to find the most pertinent characteristics. The TCN model, designed to capture short- and long-term EEG patterns, is trained using stochastic gradient descent (SGD) with a set learning rate. Pre-processed signals and selected features serve as inputs, while labels, indicating pre-ictal or ictal states, form the output. The whole approach is written in Python and leading machine learning frameworks, including TensorFlow and PyTorch. The final model is deployed as a web application as a Flask application so that clinicians and researchers can upload the EEG signals and get real-time predictions. The accuracy and AUC-ROC of the suggested TCN model were 88.7% and 90%, respectively.