Predicting food craving in everyday life through smartphone-derived sensor and usage data

Area of Science:

• Digital Health
• Behavioral Science
• Human-Computer Interaction

Background:

• Food cravings are linked to unhealthy eating behaviors like overeating and binge eating.
• Cravings fluctuate daily and are influenced by various internal and external contexts.
• Predicting food cravings can facilitate proactive digital interventions.

Purpose of the Study:

• To determine if upcoming food cravings can be detected and predicted using passive smartphone sensor data.
• To assess the feasibility of predicting cravings without requiring frequent participant questionnaires.

Main Methods:

• Collected momentary food craving ratings six times daily for 14 days from 56 participants.
• Utilized passive smartphone sensor data (noise, light, movement, screen activity, notifications, time of day) recorded prior to craving ratings.
• Excluded geolocation data from the predictor variables.

Main Results:

• Individual high vs. low craving ratings were predicted with a mean area under the curve (AUC) of 0.78.
• This predictive model outperformed a baseline model using past craving data by 14% for 85% of participants.
• The achieved AUC is considered an upper bound, requiring validation with larger datasets for training, validation, and testing.

Conclusions:

• Food craving states can be forecasted by analyzing external and internal circumstances captured via smartphone sensors and usage patterns.
• This predictive capability enables just-in-time adaptive interventions.
• Passive data collection minimizes participant burden, paving the way for scalable digital health solutions.