Accurate estimation of cellular network throughput is essential for effective network management and user experience optimization. This study conducts a comprehensive comparative analysis of four prominent machine learning (ML) models (i.e., Random Forest Regressor (RFR), Gaussian Process Regressor (GPR), K-Nearest Neighbors (KNN), and Support Vector Regressor) for predicting downlink and uplink data rates based exclusively on the user-base station distance. Evaluation metrics, including coefficient of determination (R2 score), Mean Squared Error (MSE), Mean Absolute Error (MAE), and computational runtime, were employed to assess model performance. Results indicate that ensemble-based (RFR) and probabilistic kernel-based (GPR) approaches outperform instance-based (KNN) and margin-based (SVR) methods in terms of predictive accuracy and error minimization. Furthermore, RFR achieves a favorable balance between accuracy and computational efficiency, making it a practical choice for real-time throughput prediction. Beyond performance estimation, the proposed approach offers potential benefits for energy optimization, enabling more efficient resource allocation both at the user device and base station (or network infrastructure) level. The findings suggest that incorporating additional network parameters and signal quality features could further improve model effectiveness in capturing the complex dynamics of cellular communications. This study employs a public real-world 4G LTE dataset comprising 135 traces (≈15 minutes each) collected under a static mobility model. The proposed distance-only prediction framework highlights that accurate throughput estimation can be achieved using a single distance feature, showing that lightweight models can remain competitive for real-time deployment.