Feature Selection for Machine Learning Based Step Length Estimation Algorithms.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human Motion Analysis

Background:

• Accurate step length estimation is crucial for applications like indoor positioning and gait analysis.
• Traditional methods rely on user-specific parameters, limiting adaptability.
• Machine learning (ML) offers a parameter-free alternative by learning from user data.

Purpose of the Study:

• To develop and evaluate an ML-based step length estimation algorithm for indoor positioning.
• To implement a systematic feature selection process for optimizing ML model performance.
• To compare the proposed method against existing state-of-the-art ML approaches.

Main Methods:

• Utilized machine learning techniques for step length estimation.
• Employed a systematic feature selection algorithm to identify optimal input features from a large set.
• Trained and tested the model on data from known and unknown individuals.

Main Results:

• Achieved a mean absolute error (MAE) of 3.48 cm for known users and 4.19 cm for unknown users.
• Demonstrated superior performance compared to current state-of-the-art ML methods (MAE of 4.94 cm for known, 6.27 cm for unknown users).
• Systematic feature selection significantly improved the accuracy of step length estimation.

Conclusions:

• The proposed ML-based approach with systematic feature selection provides a highly accurate and adaptable solution for step length estimation.
• This method eliminates the need for user-specific parameter tuning, simplifying implementation for diverse users.
• The findings have significant implications for enhancing indoor positioning systems and gait analysis tools.