Medhere: A Smartwatch-based Medication Adherence Monitoring System using Machine Learning and Distributed Computing

Area of Science:

• Biomedical Engineering
• Machine Learning
• Wearable Technology

Background:

• Poor medication adherence is a significant public health issue, leading to adverse health outcomes and increased healthcare expenditures.
• Timely medication reminders and adherence monitoring are crucial for effective treatment management.
• Existing methods for monitoring medication adherence are often passive or require manual input, limiting their effectiveness.

Purpose of the Study:

• To develop and evaluate a smartwatch application for objective monitoring of medication intake actions.
• To leverage inertial sensor data and machine learning for automated detection of medication adherence behaviors.
• To improve medication adherence rates through timely reminders and data-driven insights.

Main Methods:

• Development of a smartwatch application utilizing accelerometer and gyroscope data to capture user actions during medication intake.
• Implementation of a server infrastructure for data collection and preprocessing of high-frequency sensor data.
• Application of Apache Spark for data distribution and a Random Forest machine learning algorithm for action prediction.

Main Results:

• Successful preprocessing of high-frequency inertial sensor data.
• High precision and recall achieved in predicting discrete actions, including medication intake.
• Demonstrated feasibility of using wearable sensor data for medication adherence monitoring.

Conclusions:

• A smartwatch application integrated with machine learning can accurately monitor medication intake.
• This technology offers a promising approach to improve medication adherence and reduce associated healthcare costs.
• Future work could involve real-time feedback and personalized intervention strategies based on adherence patterns.