Weighted Visibility Graph With Complex Network Features in the Detection of Epilepsy

Their data

1. five data set, 100 single channel of EEG signals, each channel EEG has 4097 data point.
2. to reduce the computation time, they segment each channel with 1024 data sample points per segment.
   4 segment with 1024 data samples.
   Figure 2 showed the examples of time series data of five subsets.

Methodology

transformation of time series EEG signals to complex networks

1. add nodes: each data smple point is seen as a node
2. add edges with direction: visibility properties.
3. add weight: 根据边与水平线的夹角.

$$ omega_ab = arctanfrac{x_tb - x_ta}{tb - ta}, a < b $$

This paper showed two tables with nodes examples and edge examples, respectively.

Feature extraction

The feature extraction process compresses the large volume EEG data into relevant and important feature vector set at the cost of minimum loss of information.

1. In this paper, we have extracted two statistical properties of network named as modularity and the average weighted degree of network as features from the weighted visibility graph as these features are able to focus on how the valuable information about the time series can be acquired by analysis the structural pattern of complex networks.

Classification

1. use two classifier: SVM and KNN classifier by using Euclidean distance.
2. 这是有监督的学习,而能源的数据是没有分类的.

Performance evaluation

true positive, true negative, false positive, false negative.

Experiments

4097 data points, 4 segments. they investigated that there is not much difference in the accuracy performance when considering segmented and non-segmented approach of EEG signal. Then there is not much difference in the accuracy performance when considering segmented and non-segmented approach of EEG signa