An Advanced Self-Similarity Measure: Average of Level-Pairwise Hurst Exponent Estimates (ALPHEE)

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Natural processes often exhibit self-similarity, quantified by the Hurst exponent.
• Wavelet transforms (WT) estimate self-similarity but are sensitive to noise and assumptions.
• Accurate gait analysis is crucial for identifying elderly fall risks.

Purpose of the Study:

• Introduce a novel method (ALPHEE) for robust Hurst exponent estimation.
• Apply ALPHEE to gait data for enhanced fall detection.
• Evaluate the impact of self-similarity features on machine learning classification.

Main Methods:

• Developed ALPHEE by integrating fractional Brownian motion (fBm) with wavelet coefficient distributions.
• Combined Hurst exponent estimates from multiple wavelet decomposition levels.
• Analyzed linear acceleration (LA) and angular velocity (AV) data from 147 elderly subjects (fallers and non-fallers).

Main Results:

• Fallers exhibited higher regularity in LA and AV signals.
• Machine learning models incorporating ALPHEE-derived self-similarity features achieved 89.65% accuracy.
• This accuracy surpasses the standard method (82.75%) and prior studies on the same dataset.

Conclusions:

• The ALPHEE method provides a more precise measure of self-similarity in gait data.
• Self-similarity features significantly improve the detection of elderly fallers.
• This approach offers a promising tool for fall prevention strategies.