Abstract

Deaths in the US have drastically increased over the past decade due to addictive behaviors and drugs. According to the World Health Organization (WHO), 1 in 20 adults between the age of 15 and 64 years are addicted to at least one illicit drug; globally, 29 million people are suffering from drug use disorder. The addiction of narcotics alters a person’s primary function as well as critical areas of the brain due to multiple reasons like genetics, hereditary, stress or pressure, and mental health conditions. It not only affects an individual but also their families. Intensive research has been launched all over the world to spread awareness about how to prevent addiction. The current problem for efficiently managing and treating these addicted individuals is the lack of biomarker for detecting cravings. If clinicians could identify cravings in individuals, they might able to design appropriate intervention strategies, including mobile based mindfulness techniques, dialectical behavioral therapy (DBT) based exercises, or direct contact with support persons to mitigate risky situations (cravings) that could otherwise result in relapse. In our work, we explored the possibility of employing wearable biosensors along with machine learning approaches to define a reliable biomarker of craving.

In this work, participants wore wrist-mounted biosensors on their non-dominant arm for all waking hours for a four-day period. An event marker was used to denote any time they perceived drug craving. For analysis, raw accelerometer data in three axes (x, y, and z) evaluated 20 minutes before and 20 minutes after each marked event. A sliding window technique with signal processing Hilbert transformation approach was applied to extract relevant features mean, variance, shape, scale, and 𝐷𝑘 (a distance measure derived using six parameters in a hypothetical six-dimensional space). These features employed in machine learning approach with two different quadratic (non-linear) models to detect cravings. The collaborative work of two machine learning models provided us an accuracy of 72% in the detection of cravings.

Date of publication

Spring 5-9-2018

Document Type

Thesis

Language

english

Persistent identifier

http://hdl.handle.net/10950/1158

Committee members

Premananda Indic, Ph.D, Sarah Sass, Ph.D, David Beams, Ph.D.

Degree

Masters in Electrical Engineering

Share

COinS