A Systematic Comparison of Muscle Deformation Sensor Placement for Subject-Independent Cognitive State Classification

The proliferation of remote work has highlighted a critical gap in digital communication: the absence of non-verbal and embodied cues, which can impede mutual understanding and productivity. This work addresses this challenge by systematically investigating an optimal anatomical site for cognitive-state estimation using non-invasive optical muscle-deformation sensors to classify states such as concentration and relaxation. To model and distinguish these muscle-deformation patterns, a Support Vector Machine (SVM) classifier was employed under a subject-independent evaluation protocol. A controlled four-site comparison across calf (gastrocnemius), trapezius, upper arm (biceps brachii), and neck (sternocleidomastoid; SCM) revealed that rather than a single best site, a trade-off was observed: calf yielded higher accuracy (), but trapezius demonstrated greater stability across participants (lower variance, standard deviation (SD) = 0.17), suggesting that optimal site selection depends on deployment context. Fine-grained confusion analysis further revealed complementary error tendencies between the two sites. For real-world deployments where robustness is paramount, trapezius is recommended; for controlled settings prioritizing peak accuracy, calf is preferable. These findings provide actionable guidance for sensor placement and motivate sensor-fusion designs that leverage complementary strengths.