GeoAI based Prediction of Eating Events in Free-Living Individuals

Keywords: machine learning, mobile health, GeoAI, human dynamics, body-worn sensors
Session Type: Virtual Paper Abstract
Day: Tuesday
Session Start / End Time: 3/1/2022 08:00 AM (Eastern Time (US & Canada)) - 3/1/2022 09:20 AM (Eastern Time (US & Canada))
Room: Virtual 41

Abstract

The mobile health (mHealth) data gathered by mobile and wearable technologies have advanced our understanding of health behavior at a fine-grained level. Space and time often play a crucial role for health behavior to take place, and this sets up an opportunity for the domain of Geospatial Artificial Intelligence (GeoAI) to contribute to designing intervention or treatment strategies for improving individual health. GeoAI-based mHealth studies can help us understand the possible roles and influence of location-based exposure, mobility patterns, and physical activity bouts on health behavior and health outcome. Here we present the extension of our previous proof-of-concept work, where multiple machine learning algorithms were evaluated for predicting eating in restaurant and food purchasing events using a sub-sample dataset from 81 study participants. Increased to 216 free-living individuals with diverse ages and demographic backgrounds in this study, we applied the Gradient Boosting algorithm with engineered spatial and temporal features from data collected by accelerometer, global positioning system (GPS), and body-worn/front-facing cameras (SenseCam) for the same prediction tasks. Our efforts in addressing the class imbalance issue will also be discussed. The goal of this study is to capture the spatial, temporal, and contextual environment that leads up to eating and food purchasing events, and findings can contribute to future designs of customized GeoAI-based mHealth interventions, such as Just-in-Time Adaptive Interventions (JITAI) for eating-related behavior.