# A novel epileptic seizure prediction model based on Cox-Stuart and Optuna

## TL;DR

This paper introduces a new model for predicting epileptic seizures using EEG data, combining machine learning techniques to improve accuracy.

## Contribution

A novel seizure prediction model using Cox-Stuart and Optuna with CNN-BiLSTM for multi-patient and individual data.

## Key findings

- The multi-patient model achieved a test accuracy of 0.9992, sensitivity of 0.9996, and specificity of 0.9988.
- The individual patient model reached an average test accuracy of 0.9996, sensitivity of 0.9995, and specificity of 1.0000.

## Abstract

In order to more accurately predict whether patients with intractable epilepsy are about to develop seizures, this paper proposes an epilepsy prediction model.

When the amount of targeted patient data is small, A Cox-Stuart and Convolutional Neural Network and Bi-directional Long Short-Term Memory (Cox-Stuart-CNN-BiLSTM) model based on multi-patient epilepsy prediction is proposed, which aims to capture common features of epileptic seizures by integrating EEG signal data from multiple patients to train the model. When there is enough data for targeted patient, an Optuna and Convolutional Neural Network and Bi-directional Long Short-Term Memory (Optuna-CNN-BiLSTM) model based on independent patient epilepsy prediction is proposed, which can train the model for EEG data of individual patients, aiming to better match physiological characteristics and seizure patterns of targeted patient.

The accuracy of the test set for multi-patient is 0.9992, the sensitivity is 0.9996, and the specificity is 0.9988; the average accuracy of the test set for independent patient is 0.9996, the sensitivity is 0.9995, and the specificity is 1.0000.

It can be proved that the method proposed in this paper has good experimental results.