Explainable AI and Machine Learning Towards Human Gait Deterioration Analysis

Explainability bridges clinical practice and AI: The researchers successfully connected machine learning predictions to clinically observable gait features that doctors already use, making the AI system's decisions interpretable and trustworthy for healthcare professionals rather than a "black box."

High accuracy with real-world validation: The study achieved 98% classification accuracy for detecting Parkinson's disease severity from gait data, demonstrating that explainable AI methods can be both accurate and transparent—addressing a key concern in responsible AI deployment in healthcare.

Cognitive decline leaves measurable physical traces: The research identified that cognitive deterioration in Parkinson's disease patients produces detectable changes in gait patterns, suggesting that non-invasive gait analysis could become a practical tool for monitoring disease progression without additional invasive testing.

Identifying "weaker features" improves model reliability: By using Layer Wise Relevance Propagation to identify which gait measurements were less important to predictions, the researchers demonstrated a method for understanding and potentially improving model robustness—important for ensuring AI systems don't rely on spurious correlations.

Dual-task analysis reveals cognitive load effects: The study showed that cognitive load significantly affects gait in healthy individuals (achieving 100% accuracy in subject verification), suggesting that gait analysis could help assess cognitive function across different populations, not just disease detection.