Implementing evidential activity recognition in sensorised homes.

Area of Science:

• Computer Science
• Artificial Intelligence
• Gerontology

Background:

• Automated recognition of Activities of Daily Living (ADLs) is crucial for smart home environments supporting the elderly.
• Existing methods often rely on probabilistic approaches with limitations in representing complex sensor data and activity patterns.

Purpose of the Study:

• To present a novel process framework for generating evidential ontology networks for ADL recognition.
• To evaluate the framework's performance in a single-person occupancy smart home setting.

Main Methods:

• Development of a process framework for evidential ontology networks.
• Utilizing a 28-day sensor data set from a single-person apartment.
• Generating evidential inference networks to represent sensor evidence and activity performance.

Main Results:

• The proposed model achieved an overall class accuracy of 83.4% and a timeslice accuracy of 95.7%.
• These results surpass previous probabilistic approaches which reported 79.4% class accuracy and 94.5% timeslice accuracy on the same dataset.

Conclusions:

• The evidential ontology network framework demonstrates superior performance in recognizing ADLs compared to probabilistic methods.
• This approach offers a robust solution for enhancing smart home functionalities for elderly care.