Agu, Emmanuel O.
Petruccelli, Joseph D.
An Android smartphone app was developed to determine alcohol intoxication levels based on a user’s gait. It uses the phone's accelerometer to detect differences in gait associated with varying alcohol ingestion levels. Signal analysis found features in the time and frequency domain indicative of intoxication levels. Machine learning methods were employed with these features, trained on data from subjects, and compared by performance on a validation set. The Random Forest method was the best classifier with a success rate of 56% on the training set and 70% on the validation set. The app was distributed for user testing and included the model to be trained with the user’s new data. Accuracy improved overall with the users to 57%. As the app is used more, the accuracy is expected to increase.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.