Driver distraction is a major problem in the United States as it has caused many fatal accidents. We read an article about a tragic accident that was due to distracted driving. Three people were killed and fourteen injured when a trucker in Indiana crashed into eight cars because he was distracted reaching for his coffee mug. We realized the need to predict driver distraction before accidents occur as it clearly is a major problem. Although current car safety systems exist, they do not always predict driver distraction early enough so that the driver becomes focused.

Before we continue, let's just clarify what electroencephalography data and machine learning are. Electroencephalography is such a long word so we can just call it EEG. But what is it? Our brains have tiny cells called neurons. When these cells get excited they release electrical signals. We used a Muse 2 headband to measure these EEG signals. Now what about machine learning, or ML for short? You might think that ML is a concept far out of reach, we used to think that! But in reality the basic concept is not too hard to grasp! There are two phases in machine learning. The first phase is the training phase in which a computer is trained to convert input data to an output. The second phase is the testing phase in which the computer now can take new input data that it has never seen before and predict the output.

To collect EEG data in a relatively cheap but effective way, we used a Muse 2 headband. This device has four electrodes meaning that it records brain activity from four unique locations on the scalp. We used scikit learn libraries to train and test the machine learning models. Additionally, we used MatPlotLib which is a Python library which assists in creating plots and diagrams to visualize results.

In our case, our machine learning model was trained to use EEG data from a driver to predict a level of distraction. For practical reasons, it is not safe to test the situation of distracting a driver in a real life situation. Instead, we used a driving simulator game called City Car Driving. After collecting data on a few University of Washington students, we analyzed the data and created four different machine models. We found that our models were accurate with an approximate seventy percent accuracy. This essentially means that the model would correctly predict the distraction state of the driver 70% of the time.