Duration discretisation for activity recognition.

Area of Science:

• Computer Science
• Artificial Intelligence
• Human-Computer Interaction

Background:

• Activity recognition is crucial for smart environments and assistive technologies.
• Learning activities from sensor data is complex, with activity duration being a key, yet challenging, parameter.
• Directly using continuous duration values in models can be difficult and less effective.

Purpose of the Study:

• To discretize activity durations using various clustering algorithms.
• To develop a probabilistic model that predicts activities and individuals based on sensor data, time, and discrete durations.
• To evaluate the impact of duration discretization on activity prediction performance.

Main Methods:

• Exploration of clustering algorithms, from visual inspection to model-based clustering, for duration discretization.
• Development of a probabilistic model integrating sensor sequences, time, and discrete duration values.
• Comparative analysis of different models based on their activity prediction accuracy.

Main Results:

• Discretizing activity durations using clustering algorithms was explored.
• A probabilistic model was successfully built incorporating discrete duration values.
• Incorporating duration information consistently improved prediction performance across different clustering methods.

Conclusions:

• Discretization of activity durations using clustering is a viable approach.
• Integrating discrete duration information significantly enhances activity recognition model performance.
• The study demonstrates a nearly 3% improvement in prediction accuracy by including duration data.