Traditionally, wearable accelerometry data has been employed for activity recognition and assessing habitual physical activity, sedentary behavior, and sleep. However, emerging research that integrates this data with machine learning (ML) modeling has demonstrated the potential of using wearable accelerometry data for digital biomarker discovery for prediction of both current and future health status and diseases. To date, the combination of ML with wearable accelerometry has led to development of models that are capable of identifying and predicting people at risk of movement disorders, cardiovascular diseases, and mental health problems.
The ability to predict health outcomes and diseases by applying ML to accelerometer data likely arises from two key factors. First, different health conditions and diseases could manifest with distinct patterns in physical behaviors. Second, advanced ML models have the capability to detect and learn these subtle changes. For example, cardiovascular diseases can lead to sleep disturbances, manifesting in accelerometry data as disruptions of circadian activity rhythms. Neurodegenerative disorders such as Alzheimer’s and Parkinson’s disease often result in changes in motor abilities and gait dynamics that may reflect in wearable accelerometry data in form of altered gait and mobility.
This project will apply ML modeling on accelerometry data to identify and predict individuals at risk of having different health problems and diseases. The objectives are:
-To investigate the extent to which ML techniques applied to accelerometry data can facilitate digital biomarker discovery and validation
-To develop diagnostic and prognostic models capable of identifying and predicting individuals at risk for specific health conditions, including movement disorders, cardiovascular diseases, and mental health problems.
-To assess the accuracy and reliability of ML models for predicting health outcomes from wearable accelerometry data
