Elderly Monitoring with Environmental Sensors

As the average age of the UK population increases, ensuring that elderly and vulnerable people can access appropriate care and support becomes increasingly important. This research investigates whether basic household sensors paired with machine learning can serve as an early warning system for carers, reducing workload while maintaining quality of care.

Working in partnership with Safehouse, this project accessed real properties outfitted with environmental sensors to develop and evaluate anomaly detection systems that can identify unusual behaviour patterns that may indicate problems requiring carer attention.

Research Objectives

Assess whether basic household sensors can effectively monitor elderly residents
Develop anomaly detection systems to identify unusual behaviour patterns
Reduce carer workload while ensuring equal or improved levels of care
Compare different time series prediction methods for environmental data

Sensor Data

The study focused on four types of environmental sensors deployed in real residential properties:

🌡️

Temperature

Room temperature monitoring

💧

Humidity

Environmental humidity levels

💡

Light

Lighting patterns and usage

🚶

Motion

Movement detection

Technical Approach

Three time series prediction methods were compared for anomaly detection:

        Methods Evaluated
        LSTM (Long Short-Term Memory): Highest accuracy, especially with combined sensor approach
ARIMA: Traditional statistical time series forecasting
Autoencoder: Recommended for real-world deployment due to processing efficiency

      

Performance was measured using:

Mean Absolute Error (MAE)
Root Mean Square Error (RMSE)
R² coefficient
Expert feedback from the Safehouse team

Key Findings

LSTM achieved highest accuracy when making predictions, particularly with a combined sensor approach
Autoencoder recommended for deployment due to faster processing time suitable for real-time systems
Combined sensor data provides better anomaly detection than individual sensors
Additional parameters needed to assist in identifying important anomalies

Industry Partnership

This research was conducted in partnership with Safehouse, providing access to two real properties equipped with environmental sensors. This collaboration ensured the research addressed genuine practical needs and received expert validation from care professionals.

Future Directions

Direct contact with carers to validate system sufficiency for their needs
Stricter monitoring protocols to correlate detected anomalies with real-world problems
Integration with carer notification systems