Epilepsy seizure prediction

参考资料中的原始数据集包含 5 个不同的文件夹，每个文件夹有 100 个文件，每个文件代表一个主题/人。每个文件是 23.6 秒的大脑活动记录。相应的时间序列被采样成 4097 个数据点。每个数据点都是不同时间点的脑电图记录值。所以我们总共有 500 个人，每个人在 23.5 秒内有 4097 个数据点。我们将每 4097 个数据点划分并打乱成 23 个块，每个块包含 178 个数据点，时间为 1 秒，每个数据点是不同时间点的 EEG 记录值。所以现在我们有 23 x 500 = 11500 条信息（行），每条信息包含 178 个数据点，持续 1 秒（列），最后一列代表标签 y {1,2,3,4,5}。第 179 列中的响应变量为 y，解释变量 X1、X2、...、X178 y 包含 178 维输入向量的类别。特别是 {1, 2, 3, 4, 5} 中的 y：5 - 睁眼，表示当他们记录大脑的 EEG 信号时，患者睁眼 4 - 闭眼，表示当他们记录 EEG 信号时患者闭上眼睛 3 - 是的，他们确定了肿瘤区域在大脑中的位置，并记录了健康大脑区域的 EEG 活动 2 - 他们记录了肿瘤所在区域的 EEG 1 - 记录癫痫发作活动 属于第 2、3、4 和 5 级的所有受试者都是没有癫痫发作的受试者。只有 1 级的受试者有癫痫发作。我们创建此数据版本的动机是通过创建数据的 .csv 版本来简化对数据的访问。虽然有 5 类，但大多数作者已经进行了二元分类，即 1 类（癫痫发作）与其余的。

The original dataset in the reference materials contains 5 distinct folders, each holding 100 files, where each file represents a single subject. Each file corresponds to a 23.6-second brain activity recording, and its corresponding time series is sampled into 4097 data points, with each data point being an electroencephalogram (EEG) recording value at a different time point. In total, there are 500 subjects, each with 4097 data points spanning 23.5 seconds.

We split and shuffle each 4097-data-point sequence into 23 chunks, each containing 178 data points corresponding to 1 second of EEG recording, where each data point is an EEG value at a distinct time point. As a result, we now have 23 × 500 = 11500 total samples (rows). Each sample contains 178 data points (columns) spanning 1 second, with the final column representing the label y ∈ {1, 2, 3, 4, 5}. The response variable y is located in the 179th column, while the explanatory variables X₁, X₂, ..., X₁₇₈ form a 178-dimensional input vector for classification tasks.

Specifically, the definitions of each y value in {1, 2, 3, 4, 5} are as follows:
5 - Eyes Open: EEG signals were recorded while the patient had their eyes open
4 - Eyes Closed: EEG signals were recorded while the patient had their eyes closed
3 - Tumor Confirmed: The location of the tumor region in the brain was confirmed, and EEG activity from healthy brain regions was recorded
2 - Tumor Region EEG: EEG activity from the tumor-bearing region was recorded
1 - Seizure Activity: Seizure-related EEG activity was recorded

All subjects belonging to classes 2, 3, 4, and 5 are non-epileptic, while only class 1 subjects exhibit epileptic seizure activity.

The motivation for developing this dataset variant is to simplify data accessibility by providing a comma-separated values (.csv) formatted version of the original dataset. Although the original dataset has 5 classes, most prior studies have adopted binary classification tasks, specifically class 1 (seizure activity) versus all remaining classes.

搜集汇总

背景与挑战

背景概述

癫痫发作预测EEG数据集包含500名受试者的脑电活动记录，共11500条1秒时长的EEG片段，标注了5种状态(含癫痫发作)。该数据集常用于癫痫发作的二元分类研究。