Agu, Emmanuel O.
The goal of this MQP was to use the gyroscope and accelerometer in a smartwatch to classify a user’s BAC. We gather gyroscope and accelerometer data of 33 participants from a smartwatch while they walk at various levels of intoxication. We compute time and frequency domain features from both sensors, as well as features derived from segmenting arm movement into forward/backward swing motion. These features are then classified using various supervised machine learning methods. Random Forest was the most accurate classifier at 79.68% for a binary classification system of the [0.00 - 0.08) or [0.08+) BAC ranges. A smartwatch app is built using this model that estimates user BAC into these two bins from their gait in real time.
Worcester Polytechnic Institute
Major Qualifying Project
Access to this report is limited to members of the WPI community. Please contact a project advisor or their department to request access
Restricted-WPI community only
Your accessibility may vary due to other restrictions.