Deep Learning, Mining, and Collaborative Clustering to Identify Flexible Daily Activities Patterns.

Area of Science:

• Gerontology
• Computer Science
• Artificial Intelligence

Background:

• Monitoring daily life activities is crucial for assessing the health and wellbeing of older adults.
• Existing methods often fail to capture the inherent flexibility and variability in daily routines.

Purpose of the Study:

• To develop a novel solution for identifying flexible daily routines in older adults.
• To accommodate variations in activity order, timing, and duration.

Main Methods:

• Combines the Gap-BIDE algorithm for pattern mining with collaborative clustering (K-means and Hierarchical Clustering Agglomerative).
• Utilizes smartwatch data and a deep learning architecture (InceptionTime model) for activity detection.
• Addresses flexibility in activity sequences, optional/compulsory activities, and time spans.

Main Results:

• The proposed solution successfully identifies flexible daily routines, considering sequence, timing, and optional activities.
• The collaborative clustering approach achieved the highest coverage of monitored data at 89.63%.
• The InceptionTime model demonstrated high accuracy in detecting daily living activities.

Conclusions:

• The developed method enhances the identification of personalized daily routines for older adults by incorporating flexibility.
• This approach offers a more realistic and effective way to monitor health and detect potential care deterioration.