Electroencephalogram (EEG) is a well known, and well used method for studying brain activity, and it's possibilities have lately stretched into the car industry, were it's capabilities of detecting sleepiness in drivers are currently being put to the test. When performing EEG signal analysis on the brain, standardized signal bands exists that are characteristic to specific states of mind, such as when a driver is feeling sleepy. However, EEG as a method for studying the brain has major problems. The signal contains a lot of information that can be redundant or irrelevant, and the result is easily influenced and deviant by other parameters, that can cause incorrectness and inaccuracy in the final prediction and classification of the signal frame. One of the important methods for reducing this inaccuracy of EEG, and also reducing the computational cost of the diagnose, is feature selection. Finding key features in the signal, that can support a reliable diagnosis of a specific state of mind, is of great importance. Especially since learning systems, incorrectly predicting or interpreting a signal in the classification stage, can lead to incorrect triggering of safety features in futuristic cars, such as cruiser control. There are many existing feature selection algorithms available, and features that has been tried in different research project. The goal of this research was to help gather more accurate inputs from EEG, through an optimization study, and to increase the reliability of EEG. And by doing so, hopefully improve safety systems in cars, that in turn could help preventing sleepiness-related accidents on roads in the future. This was realized through a study of features, and feature selection algorithms. By determining key features that could distinguish sleepiness from a signal, as well as performing accuracy tests for different feature selection algorithms, the motivation for an optimal selection, based on the used parameters, could be made. However limited this research was, it concluded that Information Gain as a method for selecting features, was the most accurate algorithm, and that some features were better to use then others, such as Huguchi's fractal dimension, and the Hjorth complexity.
| Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:mdh-29852 |
| Date | January 2015 |
| Creators | Danielsson, Stefan |
| Publisher | Mälardalens högskola, Akademin för innovation, design och teknik |
| Source Sets | DiVA Archive at Upsalla University |
| Language | English |
| Detected Language | English |
| Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
| Format | application/pdf |
| Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0014 seconds