Vehicle turn-in rate is a critical metric adopted widely in highway service area (HSA) location selection, design, and service provision. Previous on-site investigations have revealed that the turn-in rate varies significantly among HSAs. Although some parametric and nonparametric learning methods have been implemented for predicting HSA vehicle turn-in rates, current research seldom tests the prediction feasibility via a random forest (RF) model. In this paper, we proposed RF models for HSA passenger and freight vehicle turn-in rate predictions. 210 HSAs turn-in rate samples are used from Sichuan, China for model calibration and validation. A comparative analysis of model prediction is further conducted via support vector regression, k-Nearest Neighbor regression, RF, XGBoost, and AdaBoost method. Compared with the rest models, the mean absolute error of the RF model on passenger and freight vehicle turn-in rate of the validation dataset is 2.36% and 2.92% respectively, which rank first and second among the tested models. It also performs well in the other two measurements. The results indicate that the RF prediction model performs well in both passenger and freight vehicle turn-in rates estimation of HSAs.