Do platform and seller cues improve price prediction? A study on online food delivery platforms using machine learning

This study investigates value-based pricing on online food delivery platforms by examining the role of beyond-core service attributes—platform brand and Observed Restaurant Activity Duration—as observable platform- and restaurant-level cues. Drawing on signaling theory as a conceptual lens, we adopt a theory-guided predictive modeling approach to assess whether these cue-related variables improve the prediction of the final price paid per order under varying levels of order urgency. Using a dataset of 1536 cross-platform transactions in Vietnam and applying a broader model-benchmarking strategy, including Random Forest, linear regression, XGBoost, and other machine-learning models, we show that nonlinear machine-learning models provide stronger out-of-sample predictive accuracy than traditional linear estimation. XGBoost achieves the strongest overall predictive performance and is therefore used as the primary model for permutation feature-importance and hypothesis analyses, while Random Forest serves as a robustness benchmark. Both platform brand and restaurant activity duration emerge as meaningful predictive contributors beyond core service attributes. Importantly, platform brand shows a substantial increase in predictive importance under high-urgency conditions, whereas restaurant activity duration remains comparatively stable across urgency levels. These findings contribute to value-based pricing and digital platform research by highlighting how theoretically informed observable cues contribute to price prediction across transaction conditions, and provide actionable insights for restaurants seeking to support pricing decisions in multi-platform environments.