Model accuracy of vision-based proctoring systems can vary across demographics, lighting conditions, and hardware setups, potentially disadvantaging some students¶

Claim · All claims
Evidence · 1 study · review 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? Reviewed studies acknowledged accuracy variation across demographics, lighting, and hardware, with potential disadvantage to certain student populations. → Malhotra 2026

Evidence¶

Malhotra 2026¶

Malhotra, M., & Chhabra, I. (2026). Ensuring academic integrity through automated online exam proctoring a decade long systematic review. Discover Education. https://doi.org/10.1007/s44217-026-01224-3

q2 · i? · review · r2

Ethical-assessment section of the review: "Several studies acknowledged that model accuracy can vary across demographics, lighting conditions, and hardware setups." The review also notes performance drops in real-world settings with varied lighting, cultural behaviors, or physical settings.

"Several studies acknowledged that model accuracy can vary across demographics, lighting conditions, and hardware setups, potentially disadvantaging certain student populations."

Discussion¶

Related Claims¶

• Head-pose-estimation-based proctoring systems report widely varying accuracies, from 75.6% for cheating detection to 100% for yaw-angle estimation — a narrower finding that bears on this claim
• Facial recognition models for proctoring report accuracies up to 99.21%, with limitations in dim-light and blurring conditions — a narrower finding that bears on this claim
• Eye-gaze-tracking-based proctoring systems report cheating-detection accuracies around 95–98% in their original studies — a narrower finding that bears on this claim
• Generative AI systems carry strong potential for introduced and inherited bias — related
• Heterogeneity within demographic categories means group variables as predictors can disadvantage atypical group members and underrepresented groups — related
• Students developed AI understanding by reasoning about how physical setup and environmental conditions affect sensor perception and machine learning data quality — related
• Infrastructure inequities — chronic bandwidth limitations, power outages, and device incompatibility — severely disrupt proctored assessments and exacerbate educational disadvantages in low-resource settings — related
• Fairness of AI outputs remains under-evaluated in edtech research — a broader claim this one bears on
• AI-based proctoring systems trained on western behavioural norms may exacerbate cultural and social challenges in non-English-speaking regions — related
• AI-detection tools produce inequitable outcomes, disadvantaging non-native English speakers through false positives — related
• Deep learning techniques such as CNNs and RNNs better detect cheating from visual cues than traditional techniques, but each carries stated trade-offs — a narrower finding that bears on this claim