This study examined the feasibility of an artificial intelligence system using webcam eye tracking to estimate sustained attention performance. In a controlled laboratory study, 129 participants (mean age: 25.98 ± 2.07 years; 83.7% female) completed the AX Continuous Performance Task. Performance was operationalized as a reciprocal-log index of correct-response reaction time, with higher values indicating faster responses. Six visual scanning features were captured, and five composite models were compared: one literature-based and four derived from experimental data. All showed significant in-sample associations with performance. The weighted composite incorporating all six features provided the best in-sample fit, but participant-level split-half cross-validation yielded a modest out-of-sample association (r = 0.19). Univariate analyses linked better performance to faster jumps and lower length and speed irregularity. These findings provide preliminary evidence for webcam-derived gaze features in cognitive assessment under laboratory conditions, with further validation needed before integration into human–computer interaction applications.
Eyes on Attention: AI-Based Webcam Assessment of Sustained Attention / Tosti, A.E., Gursesli, M.C., Duradoni, M., Colombini, G., Uccheddu, M., Thawonmas, R., Lanata, A., Khan, I., Fernández Galeote, D., Hamari, J., Sousa, C., Guazzini, A.. - In: INTERNATIONAL JOURNAL OF HUMAN-COMPUTER INTERACTION. - ISSN 1044-7318. - ELETTRONICO. - (2026), pp. 1-25. [10.1080/10447318.2026.2751375]
Eyes on Attention: AI-Based Webcam Assessment of Sustained Attention
Tosti, Anna Enrica;Gursesli, Mustafa Can;Duradoni, Mirko;Colombini, Giulia;Lanata, Antonio;Guazzini, Andrea
2026
Abstract
This study examined the feasibility of an artificial intelligence system using webcam eye tracking to estimate sustained attention performance. In a controlled laboratory study, 129 participants (mean age: 25.98 ± 2.07 years; 83.7% female) completed the AX Continuous Performance Task. Performance was operationalized as a reciprocal-log index of correct-response reaction time, with higher values indicating faster responses. Six visual scanning features were captured, and five composite models were compared: one literature-based and four derived from experimental data. All showed significant in-sample associations with performance. The weighted composite incorporating all six features provided the best in-sample fit, but participant-level split-half cross-validation yielded a modest out-of-sample association (r = 0.19). Univariate analyses linked better performance to faster jumps and lower length and speed irregularity. These findings provide preliminary evidence for webcam-derived gaze features in cognitive assessment under laboratory conditions, with further validation needed before integration into human–computer interaction applications.| File | Dimensione | Formato | |
|---|---|---|---|
|
Eyes on Attention AI-Based Webcam Assessment of Sustained Attention.pdf
accesso aperto
Tipologia:
Pdf editoriale (Version of record)
Licenza:
Open Access
Dimensione
2.23 MB
Formato
Adobe PDF
|
2.23 MB | Adobe PDF |
I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.



