Human Gait Analysis and Prediction Using the Levenberg-Marquardt Method.

Area of Science:

• Biomechanics
• Biomedical Engineering
• Machine Learning

Background:

• Accurate gait data prediction is crucial for developing effective lower limb prosthetics and orthotics.
• Existing methods may lack the precision required for personalized device design and rehabilitation analysis.

Purpose of the Study:

• To develop a high-accuracy neural network model for predicting future gait angles at fixed speeds.
• To assess the model's efficacy in gait data validation and comparison for rehabilitation engineering applications.

Main Methods:

• Gait data collected using a Biometrics goniometer from subjects walking on a treadmill at various speeds (2.4, 3.6, 5.4 kmph).
• Data pre-processed in Matlab, including filtering for movement artifacts.
• A neural network designed using the Levenberg-Marquardt method, with data split for training (60%), validation (20%), and testing (20%).

Main Results:

• The neural network achieved a mean-squared error of 10^-3 or lower, indicating high accuracy.
• Pearson's correlation coefficient and correlation plots confirmed strong agreement between predicted and actual gait data for untrained inputs.

Conclusions:

• The developed framework successfully predicts gait data, offering a viable tool for designing custom lower limb prosthetics and orthotics.
• The model can be utilized for validating gait data and comparing it against expected parameters in rehabilitation engineering settings.