Accessing Adolescent Neurotransmission through Multivariate Modeling

Overview

Adolescence (ages 14–18) is characterized by rapid neurodevelopment that influences attention and cognitive control, often producing measurable changes in reaction time and accuracy. However, the extent to which modifiable daily habits contribute to variability in cognitive control during this period remains unclear. Utilizing multivariate modeling, this study aimed to observe whether Stroop-task performance, a proxy for cognitive efficiency, could be predicted through a mix of both biological and environmental factors. Sixty-four high school students (16 per grade) were structurally surveyed on both categories, such as average sleep duration, stress (objective scale), caffeine intake, primary activity type/frequency, and demographics, then performed a computerized Stroop test; mean reaction time and accuracy across trials were calculated. The results show that these self-reported variables are surprisingly predictive. While linear regression and a two-layer MLP neural network both performed well, the neural network captured non-linear trends, yielding high validation values and showing the interconnectedness of the factors. SHAP analyses revealed that sleep and structured activity were the strongest predictors of success, while high stress and heavy caffeine intake were consistently tied to diminished performance. These findings indicate that interpretable machine learning can move beyond prediction to illuminate specific, modifiable behavioral factors linked to adolescent cognitive performance, offering potential insight for targeted interventions.