EEG-Based Alcohol Detection System for Driver Monitoring

Abstract

Today, alcohol drinking frequently accompanies socialising as a routine activity in various groups of society. 84.0% of individuals aged 18 and above in the United States have drunk alcohol at some point in their life (National Institute on Alcohol Abuse & US, 2023). Similarly, 81.7% of Norwegians in the age group 16 to 79 have drunk alcohol in 2021 (Bye, 2018). Driving after the consumption of alcohol is a worldwide problem, causing a large number of deaths and injuries a year. This work proposes the first steps towards developing an electroencephalography (EEG)-based alcohol detector conceived with the idea to prevent people from driving under the influence of alcohol. This includes the design of an experimental protocol for EEG data collection, during which participants performed the Flanker task, and their blood alcohol concentration (BAC) was measured. The resulting data set consists of two sessions per participant, both while they are affected and not-affected by alcohol. Statistical analysis of the Flanker task indicated that participants were affected by alcohol and, therefore, their EEG signals were expected to be affected as well. The collected EEG signals were used as input for intra-subject and inter-subject models, both based on the EEGNet architecture. The intra-subject model obtained a mean classification accuracy of 90.7% and the inter-subject model a mean classification accuracy of 62.9%. The result suggest that alcohol can be detected with high accuracy when developing individual models and above the change accuracy when using a general model. Therefore, the work presented here could be used as the first steps towards the development of an EEG-based alcohol detector for drivers.

Keywords

• Electroencephalography (EEG)
• alcohol detection
• Convolutional Neural Network (CNN)
• EEGNet
• Flanker Test