Neurobiosensory and Visual Patterns of Attentional and Emotional Functioning in Children Using Machine Learning:...

Attentional functioning in childhood emerges from the interaction between cognitive, behavioral, and emotional processes. Children with these vulnerabilities frequently exhibit executive function deficits, including working memory, inhibitory control, cognitive flexibility, and processing speed, which compromise the ability to regulate attention, plan, and organize tasks, and modulate emotional responses, highlighting the need for screening approaches that integrate multiple dimensions of child functioning.

This study aimed to develop and evaluate an automated, real-time, and simultaneous multimodal screening approach capable of identifying individualized cognitive-emotional-attentional profiles and behavior patterns in children through the integration of oculomotor, emotional, and neuropsychological measures, supporting earlier detection and personalized interventions.

A cross-sectional design was used to assess a sample of 260 children aged 7-15 years through a multimodal approach integrating neuropsychological tests (2- and 3-symbol cancellation tasks from the Coimbra Neuropsychological Assessment Battery; Trail Making Test Parts A and B), eye-tracking metrics, and automated facial emotion recognition during rapid naming and emotion recognition tasks. Oculomotor, emotional, and cognitive data were captured simultaneously and in real time, enabling dynamic characterization of cognitive-emotional interactions at millisecond resolution during task performance. A machine learning pipeline based on a random forest classifier, with a gradient boosting surrogate model used for Shapley additive explanations interpretability analyses, analyzed these data to differentiate children with lower attentional performance from children with normative attentional performance.

The random forest classifier achieved the best performance in the exploratory 80/20 hold-out evaluation (area under the receiver operating characteristic curve=0.9527; F₂-score=0.9055; accuracy=94.2%; sensitivity=88.5%; specificity=100%). To validate the absence of test-set selection bias, a nested 5×5 stratified cross-validation was conducted with model selection based exclusively on inner-fold F₂-scores. The negligible difference between nested cross-validated area under the curve (AUC; mean 0.9506, SD 0.0378) and the hold-out AUC (ΔAUC=0.0021) confirmed that the original result was not inflated. Children with lower attentional performance showed increased emotional reactivity, greater gaze instability, fragmented visual exploration patterns, higher variability in fixation behavior, and more frequent impulsive saccadic movements during emotionally salient tasks. Anger and fear stimuli showed stronger discriminative value across attentional performance levels, suggesting that emotional processing modulates attentional allocation.

The simultaneous, real-time capture of eye-tracking, emotion recognition, and neuropsychological data, combined with post hoc analysis within an interpretable machine learning framework, may support the identification of behavioral patterns associated with attentional functioning in children. This multimodal approach provides ecologically grounded and behavior-centered indicators that may support future screening and intervention strategies in educational and clinical contexts.