Social media engagement is shaped jointly by user gratifications and platform-embedded behavioural mechanisms. This study models Snapchat user engagement as both an explanatory and a predictive problem, using social gratification, habit, streak-related reinforcement, hedonic-experience indicators and self-expression as predictors. Survey data from 310 Snapchat users were analysed in two stages after measurement and data-integrity diagnostics. The social-gratification, streak and engagement scales were reliable (α = 0.790–0.922), whereas the hedonic items were not and were retained as separate indicators. An ordinary least squares model with HC3 heteroskedasticity-robust standard errors explained 76.7% of the variance in engagement (adjusted R² = 0.762). Social gratification (β = 0.310), streak reinforcement (β = 0.302), comfort while using Snapchat (β = 0.237) and self-expression (β = 0.226) were positively associated with engagement, whereas habit and enjoyment were not significant. Linear Regression, Random Forest and Gradient Boosting were then compared using ten-fold cross-validation. Both ensemble models achieved higher mean cross-validated R² than the linear model (0.827 for Gradient Boosting and 0.799 for Random Forest versus 0.739), although the difference between the two ensembles was small relative to fold-to-fold variability. Impurity-based and permutation importance consistently ranked streak and comfort highest and habit lowest. The study shows how business-analytics methods can complement conventional explanatory analysis by separating statistical association from out-of-sample predictive performance.