期刊
EXPERT SYSTEMS WITH APPLICATIONS
卷 217, 期 -, 页码 -出版社
PERGAMON-ELSEVIER SCIENCE LTD
DOI: 10.1016/j.eswa.2022.119488
关键词
Electroencephalography; Evolutionary feature selection; Genetic algorithm; K-means clustering; Support vector machine
In this study, a Genetic Algorithm (GA) was proposed for feature selection in EEG signals. Three different fitness functions were used without relying on expert knowledge. The results showed that the proposed method outperformed benchmarking techniques, especially in handling heterogeneous data and reducing feature dimensionality. Future works will focus on the optimization of GA parameters and the hybrid use of the two approaches.
Overview: The electroencephalographic (EEG) signals provide highly informative data on brain activities and functions. Therefore, it is possible to extract a great variety of features from these data. Problem: The heterogeneity and high dimensionality of the EEG signals may represent an obstacle for data interpretation. The introduction of a priori knowledge has been widely employed to mitigate high dimensionality problems, even though it could lose some information and patterns present in the data. Moreover, data heterogeneity remains an open issue that often makes generalization difficult.Methods: In this study, we propose the adoption of a Genetic Algorithm (GA) for feature selection, where we introduced a series of modifications on the stopping criteria and fitness functions only and that can be used with a supervised or unsupervised approach. Our proposal considers three different fitness functions without relying on expert knowledge. Starting from two publicly available datasets on cognitive workload and motor movement/imagery, the EEG signals are processed, normalized and their features computed in the time, frequency and time-frequency domains. The feature vector selection is performed by applying our GA proposal and compared with two benchmarking techniques, i.e., using the entire feature set and reducing it through principal component analysis.Results & Conclusions: Our proposal experiments achieve better results in respect to the benchmark in terms of overall performance and feature reduction. Moreover, the application of our novel fitness function outperforms the benchmark when the two considered datasets are merged together, showing the effectiveness of our proposal on heterogeneous data. The selected features are compliant with the neuroscientific literature regarding the considered experimental conditions. Future works will focus on providing a better scoring for the unsupervised technique, the hybrid use of the two approaches and the optimization of the GA parameters.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据