Skip to main content
NUTRITION 2026 Main banner

(PTFS06-02-26) Leveraging Machine Learning and Ecological Momentary Assessment Data to Identify Factors That Predict Children's Energy Intake

Sunday, July 26, 2026
11:04 AM - 11:08 AM ET
Location: Prince George's Hall B-E (The Hub) - Poster Theater 2 / Poster Board Number: PT2-2
Emerging Leaders in Nutrition Science Poster Competition
Disclosure(s):
Hanim Ecem Diktas, PhD, MS: No relevant financial relationship(s) with ineligible companies to disclose.
Objectives: Ecological Momentary Assessment (EMA) studies show near-real-time physical, social, and contextual factors are associated with changes in children’s dietary intake. Because machine learning (ML) can utilize high-dimensional data, this study leverages EMA and ML to identify the dynamic predictors of children’s energy intake (EI).

Methods: EMA and dietary data from the Mothers And Their Children's Health (MATCH) study, 3-year intensive longitudinal, observational, dyadic design (190 mothers and their 8 to 12 year-old children), were analyzed. EMAs from mothers and children were matched to time-stamped 24-hour dietary recalls, each EMA matched to an eating occasion within a ±2-hour window. When multiple eating occasions were within a single EMA window, we prioritized the occasion after the EMA (pre-meal EMA); otherwise, the most proximate eating occasion was selected, and each occasion was matched with only one EMA. We evaluated Random Forest (RF), Decision Tree Regressors, Linear Regression, Light Gradient Boost Machine, and Extreme Gradient Boost Regressor Models using group-aware 10‑fold cross‑validation by dyad ID. The metrics for predicted EI was mean absolute error (MAE), Root Mean Squared Error (RMSE) and R². Mean absolute Shapley Additive Explanations (SHAP) values were used to interpret the best performing model.

Results: In the matched dataset, children (49% male; mean ± SD, age: 10 ± 1 y) had 3366 eating occasions (63% main meals; 37% snacks). The best-performing model, RF, achieved MAE= 207.5 ± 13.9 kcal, RMSE= 286.0 ± 20.6 kcal, and R²= 0.25 ± 0.05. SHAP values indicated that the type of eating occasion (snack vs. meal) and eating location (home vs. out-of-home) were the most influential predictors of energy intake. While snacks decreased predicted EI (48% contribution), eating out-of-home increased predicted EI (13%). Contextual factors, including screen use (1.1%) and having sports activity (1.2%) increased predicted EI but with more modest contributions.

Conclusions: This group-level model identified factors predicting children’s EI with errors comparable to prior research. Further research should test individual-specific models to examine within-individual variability and enable precision nutrition with individualized dietary recommendations tailored to physical, social, and contextual factors.

Funding Sources: NIHLBI, NICHD