Predictive performance of sensor configurations does not change monotonically with the number of sensing streams; preferred configuration depends on target metric and sensing cost¶

Claim · All claims
Evidence · 1 study · design r2 · q2
Currency · newest study 2026 · checked against its sources 2026-10-20 (automated, 1 of 1 entry) · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? Smaller sensor configurations sometimes achieve similar or better observed results than larger ones; the five-stream EEG+EDA+eSense IMU+HR+Ring Temperature configuration achieved the highest Within-1 accuracy (85.86%) and binary accuracy (74.44%), while ECG alone achieved the highest binary Macro-F1 (70.54%). → Zikang Leng 2026

Evidence¶

Zikang Leng 2026¶

Zikang Leng, Edan Eyal, Yingtian Shi, Jiaman He, Yaqi Liu, and Thomas Plötz. (2026). EduGage: A Multimodal Dataset and Benchmark for Sensor-Based Momentary Assessment of Engagement in Self-Guided Video Learning. Proc. ACM Interact. Mob. Wearable Ubiquitous Ubiquitous Technol. https://arxiv.org/abs/2605.01238 doi:10.48550/arxiv.2605.01238

q2 · i? · design · r2

Modality configuration tradeoff study screening all 2,047 nonempty combinations of the 11 sensing streams under the same participant-grouped folds (Table 5). Reduced configurations such as the five-stream EEG, EDA, eSense IMU, HR, and Ring Temperature setup reached the highest Within-1 accuracy (85.86±4.72%).

"Performance does not change monotonically with the number of streams: smaller configurations sometimes achieve similar or better observed results than larger ones."

Discussion¶

Related Claims¶

• Chest-worn ECG provides the strongest cardiac-sensing performance for attention-difficulty prediction, and fusing all cardiac streams does not improve upon chest ECG alone — related
• A head-confined multimodal configuration (EEG, eye tracking, forehead EDA) predicts held-out learners' five-level engagement ratings at 75.0% balanced 1-off accuracy, 12.0 points above the mode baseline — related
• A modality-aware reference model outperforms sensor-free, statistical, deep temporal, foundation-model, and LLM-based baselines for estimating minute-level attention difficulty from wearable sensing — related
• Larger Llama Guard models outperform smaller ones on education-prompt classification, with the 8B model best in accuracy, recall, and F1 — reports the opposite
• A tool's effectiveness results from the whole configuration of events, activities, and contexts in which it is used — related
• Combining more modalities does not always improve detection of learner mental states — a broader claim this one bears on
• Model-family rankings depend on the prediction target: AdaBoost leads the five-level ordinal task while LightGBM holds the highest binary macro-F1 — a narrower finding that bears on this claim