Faculty Advisor

Agu, Emmanuel O.

Abstract

Alcohol abuse has been a pervasive problem worldwide, causing 88,000 annual deaths. Recently, several projects have attempted to estimate a user’s level of intoxication by measuring gait using mobile sensors. The goal of this project was to compare a deep learning approach to previous methods to predict the blood alcohol concentration of a user by training a convolutional neural network and creating a mobile app which could accurately determine intoxication level. We gathered data from 38 participants over the course of 12 weeks, collecting accelerometer and gyroscope data simultaneously from both a smartwatch and smartphone. Our neural network’s accuracy is roughly 64% on the test set and 69% on the training set into 5 BAC ranges for an input containing two seconds of data.

Publisher

Worcester Polytechnic Institute

Date Accepted

March 2018

Major

Computer Science

Project Type

Major Qualifying Project

Accessibility

Unrestricted

Advisor Department

Computer Science

Share

COinS