A Proposed Explainable AI Framework for Early Detection of Developmental Risks Associated with Excessive Screen Exposure in Toddlers

Abstract

The rapid increase in digital media exposure among young children has raised concerns regarding its potential impact on cognitive, linguistic, behavioral, and socio-emotional development. While previous studies have reported associations between excessive screen exposure and developmental challenges, most existing approaches rely on simple duration-based measures and provide limited insight into the factors contributing to individual risk. Furthermore, conventional machine learning models often operate as "black boxes," making their predictions difficult for parents, educators, and healthcare professionals to interpret. This paper proposes an Explainable Artificial Intelligence (XAI) framework for the early detection of developmental risks associated with excessive screen exposure in toddlers. The framework utilizes behavioral, environmental, and screen-usage indicators, including daily screen duration, bedtime device use, co-viewing patterns, caregiver interaction, reading frequency, sleep regularity, and outdoor play engagement. Multiple supervised learning algorithms, including Random Forest, XGBoost, and LightGBM, are employed for risk classification. To enhance transparency and trustworthiness, SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations) techniques are integrated to identify and explain the most influential factors affecting model predictions. The proposed framework aims not only to predict developmental risk levels but also to provide interpretable insights that can support informed decision-making and early intervention strategies. The study contributes to the growing field of pediatric digital health by combining predictive analytics with explainable machine learning for child well-being assessment.

Keywords