Answers are grounded in this paper and cite their sources.
Demonstrates eye rubbing detection using smartwatch technology, suggesting improved ocular health monitoring.
Frequent eye rubbing (ER) is associated with progressive ocular conditions such as keratoconus and dry eye disease. Continuous and objective ER monitoring is therefore clinically important, yet remains challenging due to class imbalance and sensitivity to hyperparameter tuning. This study proposes a smartwatch-based ER detection framework built upon an imbalanced area under the curve-optimized proximal support vector machine (ImAUC-PSVM). The proposed classifier integrates direct AUC maximization into the PSVM objective function. This modification improves discrimination between minority and majority classes. Theoretical analysis shows that the reformulated objective preserves the convex structure and the closed-form solution of the standard PSVM. At the same time, it introduces an AUC-driven loss that improves robustness under skewed distributions and enables incremental learning. To further stabilize performance, Bayesian Optimization Hyperband (BOHB) is employed for efficient hyperparameter tuning. The framework was evaluated on the Hand-Face Interaction Dataset (HFID) dataset using stratified 5-fold cross-validation. The proposed model achieved 92.093% accuracy, 90.357% F-measure, 91.694% G-means, and an AUC of 0.815. When compared with the strongest deep learning baseline, the proposed model improved accuracy by 2.37%, F-measure by 1.71%, and AUC by 0.025. These results establish the potential for the method to serve not only as a robust tool but also as a real-time tool that can operate efficiently in clinical or wearable device settings.
No takes yet. Share an insight, caveat, or question.
Wu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: