Investigating the Impact of Fast-Paced Digital Content, Social Media Engagement, and Lifestyle Factors on Attention,...

The increase in fast-paced digital content and changes in lifestyle among young adults have raised concerns about their effects on cognitive functions. While many studies have looked into this connection, few have used strong machine learning models to predict subclinical cognitive outcomes based on real-world, self-reported behaviors.

This study aimed to develop and evaluate machine learning models that predict levels of attention, cognitive load, and behavioral impulsivity using self-reported digital usage and lifestyle factors.

A cross-sectional study included 301 participants who filled out a custom 55-item online questionnaire. Four machine learning algorithms (K-nearest neighbors [KNN], support vector regression [SVR], random forest, extreme gradient boosting [XGBoost]) were evaluated for each of the three cognitive-behavioral outcomes using a 5 × 5 nested cross-validation for unbiased evaluation. Feature importance was established using permutation importance and confirmed with SHapley Additive exPlanations (SHAP).

The models showed strong predictive power, especially for attention, where the random forest model had the best performance (R² = 0.513 ± 0.075). SVR was most effective for cognitive load, while XGBoost was the best for behavioral impulsivity. A key finding from the feature importance analysis was that technology usage habits were the main predictors for attention. In contrast, lifestyle factors were the strongest predictors for both cognitive load and impulsivity.

Machine learning models can reliably predict cognitive and behavioral outcomes from easily accessible self-reported data. Findings indicate that inattention is most strongly associated with digital usage behaviors, whereas cognitive load and impulsivity are more strongly associated with indicators of underlying emotional distress; because the design is cross-sectional, these relationships are predictive rather than causal.