الفهرس 
Introduction and MotivationOnly 14 pages are availabe for public view
 |  | from | 112 |  |  |
| | | from | 112 | | |
Abstract
Traffic accidents are one of the major global concerns that claim millions of lives annually in addition to their major economic consequences. It is well-recognized that it has become one of the biggest threats to human life.
Despite extensive efforts to reduce and control this issue, daily statistics show that road traffic accidents are on the rise. It is recorded as the world’s ninth cause of death worldwide, even after the covid-19 pandemic, and is predicted to be the fifth by 2030.
With emergency braking support systems, a considerable percentage of car accidents could be avoided. Over the last decade, many studies investigated the possibility of using the driver’s brain signals to observe and predict the driver’s behavior, distraction, fatigue, alertness, lane-changing intention, and emergency braking intention. These researchers aim, first, to help disabled individuals drive in a normal and more controlled way. Secondly, to decrease the percentage of accidents and even someday prevent them altogether.
In this thesis, we propose a model that can predict the driver’s intention to perform emergency braking using electroencephalograph (EEG) brain signals combined with leg electromyography (EMG). The data set used for this study contains the EEG+EMG signals of 18 subjects collected while driving a simulated car with an electrode cap. The EEG signals are segmented into 150ms windows of signals after leaving six different time gaps [150ms, 140ms, 130ms, 120ms, 110ms, and 100ms] from the braking action. Three feature extraction methods are investigated on EEG signals (time domain, frequency domain, and discrete wavelet transform). For the classification phase, we compared five classifiers (K-Nearest Neighbors, Logistic Regression, Random Forest, Naïve Bayes, And Support Vector Machine). The Random Forest, K-Nearest Neighbors, and support vector machine achieved high-performance with an average accuracy of 99.7% when combined with discrete wavelet transform feature extraction method 150ms before the braking instance. In other words, the proposed model can predict the driver’s intention to perform emergency braking action 4.22m earlier at 100 km/h driving speed.