Autumn's Dawn NICE Laboratory: Downloads
Human Brain Mapping 2011 Conference Posters
- Data Driven Connectivity Changes in Patients with Alzheimers
- Connectivity in math−gifted adolescents: Comparing Structural Equation Modeling with Granger Causality Analysis
NIWeek 2011
Feature Selection Toolbox
Installation Instructions
In order to install this program EEGLab must be installed. You can download EEGLab from http://sccn.ucsd.edu/eeglab/
Extract this folder to the plugins folder in the eeglab directory. When you start EEGLab and load one or more files the tools menu is activated and should include a tab titled classification.
Usage
This program is designed to classify multiple groups of EEG. Currently the design works for any files that can be imported into EEGLab.
Step 1:
Load in all of the files for all of the classes to run the classification on.
Step 2:
In the menu at the top of the screen, follow these steps:
tools −> FST −> feature generation -> Average Power, Coherence, Correlation, or Phase Synchrony
Choose which method of feature generation to use and adjust any settings as needed. Press Ok to execute the feature generation.
note: Coherence feature generation has the ability to only generate features on certain channel sets. The sets correspond to different types of coherence as described by [2]. The imported data must include 10−10 or 10−20 standard labels for this method to work.
Step 2a (optional):
tools −> FST −> save features to Excel file
This will run feature reduction on the set of generated features. There are 3 algorithms that are all related. [3] [4].
Step 3:
tools −> FST −> Feature Reduction
The algorithms to choose from are:
- SFS − only goes forward without checking for back steps, much faster but worse OCR results
- SFFS − Backwards check is enabled to look for better group results, slower but significantly better
- IFFS − improved forward floating selection, an additional backwards check is added, slowest but slightly better OCR than SFFS
Max features is the number of features the selection algorithm runs to. When SFFS or IFFS is selected an additional value of 5 is added to this in order to insure backwards steps can be made if needed.
This can take a very long time to run on large feature sets for large max feature values.
Step 4:
tools −> FST −> save features to excel file
At this point the selected features from SFFS and their labels are saved to an excel file in sheet 1 while the overall classification rate is saved to sheet 2.
Electrical & Computer Engineering
-
Address
Texas Tech University, Box 43102, 2500 Broadway, Lubbock, TX 79409 -
Phone
806.742.3533 -
Email
webmaster.coe@ttu.edu