A strategy based on paraconsistent random forest for sEMG gesture recognition systems robust to contaminated data.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Signal Processing

Background:

• Machine learning on physical signals is challenging due to uncontrolled acquisition environments.
• Surface electromyography (sEMG) signal degradation from noise and artifacts complicates movement recognition.
• Existing pattern recognition algorithms struggle with altered myoelectric signal characteristics.

Purpose of the Study:

• To present the Paraconsistent Random Forest (PRF) method for enhanced sEMG-based movement recognition.
• To evaluate the robustness of PRF against common sEMG signal contaminants.
• To demonstrate PRF's superiority over traditional methods in degraded data scenarios.

Main Methods:

• Developed a hybrid classifier combining Random Forest's noise resilience with Paraconsistent Logic's ability to handle non-ideal data.
• Utilized experimental procedures to test the method's performance with movement artifacts, thermal noise, and electrode-skin contact loss.
• Statistically validated all experimental results.

Main Results:

• The Paraconsistent Random Forest method showed a decrease of less than 20% in movement prediction accuracy with degraded data.
• Traditional methods experienced up to 90% decrease in prediction accuracy, often becoming invalidated.
• PRF demonstrated significant robustness and viability in the presence of typical sEMG contaminants.

Conclusions:

• The Paraconsistent Random Forest method is a promising approach for machine learning applications involving degraded physical signals.
• PRF significantly improves the reliability of movement recognition from sEMG data in real-world, non-controlled conditions.
• Hybridization of Random Forest and Paraconsistent Logic enhances decision tree representational power for vague or contradictory data.