Intention Prediction and Human Health Condition Detection in Reaching Tasks with Machine Learning Techniques

Area of Science:

• Robotics
• Biomedical Engineering
• Machine Learning

Background:

• Human-robot interaction relies on accurate motion detection and intention prediction.
• Machine learning techniques (MLT) offer solutions for limited data scenarios in applications like robotic rehabilitation and industrial collaboration.
• Analyzing body signals is crucial for understanding human movement patterns.

Purpose of the Study:

• To compare the performance of Linear Discriminant Analysis (LDA) and Random Forest (RF) for human motion analysis.
• To evaluate MLT's ability to predict movement intention and detect pathological movement patterns in post-stroke patients.
• To assess the impact of signal length and arm used on prediction accuracy.

Main Methods:

• Utilized wearable electromagnetic sensors to capture reaching movement data from healthy subjects and post-stroke patients.
• Applied Linear Discriminant Analysis (LDA) and Random Forest (RF) algorithms to analyze signal sub-sections.
• Evaluated prediction accuracy for movement intention, health condition (healthy vs. pathological), and brain hemisphere damage.

Main Results:

• Accuracy improved with longer signal portions (at least 10%) and when analyzing only healthy subjects (up to 11%).
• Random Forest (RF) demonstrated superior performance in predicting movement intention (62.19% vs. 59.91% for LDA).
• RF achieved over 90% accuracy in detecting health conditions, outperforming LDA.

Conclusions:

• Machine learning techniques, particularly RF, are effective for analyzing human motion and predicting intentions in human-robot interaction.
• RF shows significant potential for real-time health condition detection and rehabilitation applications.
• Signal length and subject health status critically influence the accuracy of motion analysis.