Using decision trees to measure activities in people with stroke

Area of Science:

• Rehabilitation Medicine
• Biomedical Engineering
• Wearable Technology

Background:

• Improving community mobility is crucial for stroke survivors.
• Measuring daily physical activity is key to assessing rehabilitation effectiveness.
• Previous research demonstrated SmartShoe's ability to classify basic activities (sitting, standing, walking) in stroke patients using Artificial Neural Networks (ANN).

Purpose of the Study:

• To develop and evaluate decision tree algorithms for classifying daily activities in stroke patients using the SmartShoe system.
• To compare the accuracy of individual versus group activity classification models.
• To assess the feasibility of multi-class activity recognition beyond basic postures.

Main Methods:

• Utilized data from 12 participants with stroke wearing the SmartShoe system.
• Developed activity classification models using decision tree algorithms.
• Evaluated models for 3-class (sitting, standing, walking) and 8-class (including cycling, stairs, wheelchair use) activity recognition.

Main Results:

• For 3-class classification, individual models achieved 99.1% accuracy, and group models achieved 91.5% accuracy.
• For 8-class classification, individual models reached 97.9% accuracy, and group models achieved 80.2% accuracy.
• Demonstrated high feasibility of SmartShoe for multi-class activity recognition in stroke patients.

Conclusions:

• Decision tree algorithms provide a feasible and accurate method for classifying daily activities in stroke patients using the SmartShoe system.
• Individualized models show higher accuracy, but group models offer a viable alternative for broader application.
• SmartShoe technology holds significant potential for monitoring physical activity and informing rehabilitation strategies for stroke survivors.