The Effect of Sensor Feature Inputs on Joint Angle Prediction across Simple Movements.

Area of Science:

• Biomechanics
• Wearable Technology
• Machine Learning in Healthcare

Background:

• Wearable sensors like inertial measurement units (IMUs) are increasingly used with machine learning for human intent recognition in health.
• Limited research exists on how the number and placement of IMUs influence human movement intent prediction (HMIP) at the joint level.

Purpose of the Study:

• To analyze the impact of IMU quantity and placement on the accuracy of predicting joint angles for simple human movements.
• To determine optimal sensor configurations for maximizing machine learning prediction accuracy in HMIP.

Main Methods:

• Trained a Random Forest algorithm to predict future joint angles using various combinations of IMU input signals.
• Evaluated prediction accuracy (RMSE) for ankle, knee, and hip joints under different IMU configurations (adjacent vs. non-adjacent placements).

Main Results:

• Adding adjacent IMUs did not significantly improve joint angle prediction accuracy (e.g., ankle RMSE 1.92° vs. 3.32°).
• Including non-adjacent IMUs also failed to enhance prediction accuracy (e.g., ankle RMSE 5.35° vs. 5.55°).
• The quantity and placement of IMUs showed minimal impact on predicting future joint angles during simple movements.

Conclusions:

• The addition of IMUs, whether adjacent or non-adjacent, does not substantially improve the accuracy of predicting future joint angles.
• Current joint angle inputs are sufficient for predicting simple movements, and additional IMUs offer limited predictive benefit.