Abstract
A MetaBase sensor was used to collect accelerometer data while a set of five desk-related activities was performed corresponding to productive and distracted desk work. Using a Naïve Bayes model, these activities were categorized with up to 81% accuracy. This methodology proves that commonly worn wrist accelerometers, such as a Fitbit, can be used to differentiate between positive activities, such as writing and typing, and negative activities, such as phone use and sitting idle. This can help students or working professionals stay on task with reminders if some notification is given when negative activities are detected. This study also explores the efficacy of various features and provides a comparative study among various machine learning algorithms. Furthermore, this study aims to be able to accurately distinguish between these activities with minimum requirements, saving memory and time in data processing and allowing for cheaper components to be used.