Instrumented smart homes offer an unprecedented opportunity to unobtrusively monitor human behavior in natural environments. Additionally, they can be used to determine whether relationships exist between behavior and health changes. Here we introduce an approach to behavior change detection (BCD) that can be used to identify behavior changes that accompany health events. BCD detects changes between time periods, determines significance of the detected changes, and analyzes the nature of the changes. In the case of smart homes, sensor data is collected and labeled using activity recognition and BCD is applied to analyze behavior changes by quantifying and analyzing changes in the activity timings and durations. We demonstrate our approach using three case studies for older adults living in smart homes who experienced major health events. Our evaluation indicates that behavior changes consistent with the medical literature do occur in these cases and that the changes can be automatically detected using BCD. The proposed smart home, activity recognition, and change detection algorithms are useful data mining techniques for understanding the behavioral effects of health conditions.
