Analyzing Daily Behavioral Data for Personalized Health Management

Abstract

Emerging wearable and environmental sensor technologies provide health professionals with unprecedented capacity to continuously collect human behavior data for health monitoring and management. This enables new solutions to mitigate globally emerging health problems such as obesity. With such outburst of dynamic sensor data, it is critical that appropriate mathematical models and computational analytic methods are developed to translate the collected data into an accurate characterization of the underlying health dynamics, enabling more reliable personalized monitoring, prediction, and intervention of health status changes. However, several challenges arise in translating them effectively into personalized activity plans. Besides common analytic challenges that come from the missing values and outliers often seen in sensor behavior data, modeling the complex health dynamics with potential influence from human daily behaviors also pose significant challenges.

We address these challenges as follows: We firstly explore existing missing value imputation and outlier detection preprocessing methods. We compare these methods with a recently developed dynamic system learning method – SSMO – that learns a personalized behavior model from real-world sensor data while simultaneously estimating missing values and detecting outliers. We then focus on modeling heterogeneous dynamics to better capture health status changes under different conditions, which may lead to more effective state-dependent intervention strategies. We implement switching-state dynamic models with different complexity levels on real-world daily behavior data. Finally, we conducted evaluation experiments of these models to demonstrate the importance of modeling the dynamic heterogeneity, as well as simultaneously conducting missing value imputation and outlier detection in achieving better prediction of health status changes.