Stefan Nio

Honours thesis Research 2024

Classifying seven seizure types from raw EEG

Role
Sole researcher
Timeline
2024
Team
Supervised research, UWA
Skills
Deep learning, Time series, TensorFlow, Keras

Overview

A WaveNet and BiLSTM pipeline for multi-class seizure classification.

My honours thesis. The model reads raw EEG and names one of seven seizure types, which is a harder problem than detecting whether a seizure happened at all.

Evaluation

I split by patient, not by recording.

A model tested on recordings from patients it trained on learns the patients. Holding whole patients out makes the numbers worse and makes them mean something.

Reflection

What I would do differently now.

This section is being written.