P03-Dimension reduction of EEG feature space by using PCA
| Autor: | Vladimir Krajca, M. Piorecký, J. Štrobl |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
business.industry
Computer science Computation Dimensionality reduction Feature vector Pattern recognition Space (commercial competition) Sensory Systems Reduction (complexity) Statistical classification Neurology Dimension (vector space) Physiology (medical) Principal component analysis Neurology (clinical) Artificial intelligence business |
| Zdroj: | Clinical Neurophysiology. 129:e14-e15 |
| ISSN: | 1388-2457 |
| DOI: | 10.1016/j.clinph.2018.01.048 |
| Popis: | EEG is often used in clinical practice. The automatic classification methods are used to help an expert with understanding of the signal. We characterize signals by using features, which mathematically describe the signal in multidimensional space. High number of selected features leads to higher computation time and can worsen performance of any automatic classification algorithms. Principal component analysis (PCA) is method often used to reduce the dimension (dimension of features space in this study). The output of our work is the size of the dimension to which we can reduce the space to maintain the required percentage of information from the original data. We analyzed 24 features from program Wave-Finder, which is used in clinical practice and we used data with 49,554 segments. We kept 98% of information in the reduction to 15 dimension of feature space. For 11 dimension we kept 95%, for 8 dimension we kept 90% and for 6 dimension we kept 85% of information. We also try to reduce space to 3D, which we can visualize. In this case we kept 70% of original information. So the dimension reduction offers the possibility of facilitating classification with corrected loss of information. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect