A temporal analysis system for early detection of health changes
Metadata[+] Show full item record
To make it possible for elders to live independently at home and yet get help from health care providers when small changes in health conditions take place, smart home technologies are developed to enhance safety and monitor health conditions via noninvasive sensors and other devices. To better analyze the wealth of the activity information from various kinds of sensors to locate trends that correspond states of wellbeing, this thesis proposes a new system to build adaptive models for detecting health changes based on temporal analysis, including outlier detection, customization and adaption to new changes. Our hope is that by using more sophisticated temporal analysis method we can capture more predictive alerts and more customized alerts that can help us detect more meaningful health changes before they become big problems. Since we cannot have full access to all the embedded sensor data from TigerPlace at the moment, the system is tested using synthetic datasets which simulate gradual changes, sudden changes, changes of baseline health condition and system noise that might happen in the real-world data. Based on the experiments on the synthetic datasets, the system is proved to have the ability to adapt to gradual changes, find anomalies and spawn a new component for the GMM when there is an emerging new normal pattern. The system achieves our goals when tested on the synthetic datasets over extended period of time. We hope that by using the system in Tiger Place, it will help by detecting health changes before real health issue happens.