Using Accelerometers in Mobile Phones to Estimate Blood Alcohol Levels

Abstract

This paper studies methods for determining the blood alcohol content of individuals by using data from commodity accelerometers in mobile phones carried on person. A significant challenge is that such data is very noisy and often irregular (many large gaps) as well. This paper provides a detailed analysis of a recently released dataset of accelerometer traces and associated readings of transdermal alcohol content (TAC). It describes a set of features extracted from the raw accelerometer traces that are effective for the task of determining TAC levels. It presents results of an experimental study of regression methods that use these features to predict TAC levels from accelerometer traces as well as of classification methods that predict whether the person carrying the mobile phone has TAC levels above given thresholds.