Poster Abstract: Detecting Kratom Intoxication in Wearable Biosensor Data

In the ongoing opioid addiction crisis, users who lack access to treatment have sought novel methods to relieve withdrawal symptoms. Among these is a psychoactive plant from South-East Asia popularly known as kratom. With its spreading consumption it would be valuable to automatically detect kratom use. Although wearable biosensors have been applied to detect substance use in the past, kratom’s effects are not as well understood and can be paradoxical. In this paper, we perform supervised learning on a set of features extracted from streaming kratom data gathered from wrist-worn biosensors deployed on participants over a period of several days.We extract several time domain features, define a period of intoxication post-use based on the existing literature, and compare four classifiers based on their accuracy, sensitivity, and specificity. Our results show that kratom use can be detected with 95% accuracy using a random forest classifier in data collected from home use.