Smartphone Classifies Single-Leg Squat Into Three Levels
Researchers at Sangji and Yonsei Universities present an interpretable machine-learning framework (2026) that classifies single-leg squat (SLS) performance into three levels (good/moderate/poor) from frontal-view smartphone videos of 105 young adults. Using 17 engineered trunk/pelvis/knee features and adaptive boosting, the model achieved 0.84 accuracy, 0.85 F1, and 0.92 AUC; SHAP and LIME highlighted coordination-informed features as primary drivers, supporting clinical screening feasibility.
Key Points
- 1Demonstrates adaptive boosting classifies SLS into three levels with 0.84 accuracy on smartphone videos
- 2Highlights coordination-informed features (summated angle, knee×trunk interaction, knee-to-trunk ratio) as dominant predictors
- 3Enables transparent, instance-level explanations via SHAP and LIME for targeted rehabilitation planning
Scoring Rationale
Combines credible peer-reviewed results and actionable, interpretable smartphone workflow; novelty is incremental relative to prior markerless SLS studies.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
