LIFETRACE: A Longitudinal Multimodal Dataset on Daily Physical Activity, Well-Being, and Habits

Digital technologies and wearable devices have emerged as critical tools for developing personalized well-being/health interventions, closely supported by psychological experience sampling instruments. While existing public datasets often focus on clinical populations and controlled environments, this paper presents a comprehensive dataset integrating anthropometric and health measurements, in-the-wild longitudinal physical activity monitoring, environmental context data, socio-demographic factors, and standardized psychological well-being metrics. By monitoring relevant behavioral and health data, this dataset investigates those relevant determinants to form long-term physical activity habits. Over a 14-week period, we collected data from 88 participants. Of these, 59 followed a daily walking protocol during the first 7 weeks, with the remaining 29 participants serving as a control group, with no physical activity required. This paper presents an exploratory data analysis (EDA) to test the reliability and completeness of the collected data, together with a baseline machine learning model to assess the predictive power of included variables, relative to the well-being metrics. Moreover, by using the feature importance method, a preliminary explainability approach is applied to evaluate the soundness of predictions regarding feature contributions. Several applications are envisioned, including the impact of environmental conditions on behavior, exploring the contribution of each observed variable to habit formation, and providing a benchmark for future machine learning models in this field.