Systematic Comparison of the Influence of Different Data Preprocessing Methods on the Performance of Gait...

Area of Science:

• Biomechanics
• Motor Control
• Machine Learning in Sports Science

Background:

• Human movement analysis involves complex, non-linear interactions.
• Accurate classification of movement patterns is crucial for practitioners.
• Machine learning shows promise for enhancing movement classification performance.

Purpose of the Study:

• To systematically compare various data preprocessing techniques for gait pattern classification.
• To evaluate the impact of different preprocessing combinations on machine learning model performance.
• To provide domain-specific recommendations for robust gait analysis.

Main Methods:

• Utilized a public dataset of gait patterns from 42 healthy participants.
• Applied preprocessing steps: GRF filtering, time derivative, time normalization, data reduction, weight normalization, and data scaling.
• Compared classification performance using Support Vector Machines, Random Forest, Multi-Layer Perceptrons, and Convolutional Neural Networks.

Main Results:

• Filtering ground reaction forces (GRFs) and supervised data reduction (e.g., Principal Components Analysis) enhanced classifier prediction accuracy.
• Weight normalization and the number of data points in time normalization had minimal impact.
• Machine learning models demonstrated improved classification with optimized preprocessing.

Conclusions:

• Filtering GRFs and supervised data reduction are recommended for improved gait pattern classification.
• Standardized preprocessing can lead to more comparable and robust machine learning models.
• Findings support practical applications in human movement analysis and sports science.