This dataset accompanies the article “Depressive Symptom Predictors in Older Mexican Adults: Interaction Structures and Non‑Linear Effects from Machine Learning Explainability.” It contains baseline data from 1,252 adults aged ≥ 60 years participating in the COSFAMM cohort (Obesity, Sarcopenia, and Frailty in Older Mexican Adults), collected in Mexico City between 2014 and 2019. The dataset includes sociodemographic characteristics, chronic disease indicators, social isolation (LSNS‑6), perceived social support (MOS‑SSS), and depressive symptoms assessed with the CESD‑R. These variables were used to develop LASSO‑guided logistic regression and Random Forest models, followed by SHAP, Friedman's H‑statistic, and ALE analyses to characterize psychosocial and clinical predictors of depressive symptoms.