Accurate path-loss prediction is essential for reliable LTE network planning, particularly in complex urban propagation scenarios where signal attenuation is influenced by radio, geometrical, and environmental factors. This study proposes a Genetic Algorithm (GA)-Based Weighted Blending framework for LTE path-loss prediction using real drive-test measurements collected along the Bus-Way corridor from Jembatan Ampera to Terminal Alang-Alang Lebar in Palembang City, Indonesia, at 1800 MHz and 2100 MHz. The dataset consisted of 19,766 cleaned samples with ten radio, geometrical, and environmental input features. Four heterogeneous machine-learning regressors, namely Random Forest, XGBoost, Multilayer Perceptron (MLP), and Support Vector Regression (SVR), were individually hyperparameter-tuned and combined through GA-based ensemble weight optimization using out-of-fold predictions. The proposed framework was evaluated against empirical propagation models, individual machine-learning models, and average blending using RMSE, MAE, R², and residual diagnostics. The results show that GA-Based Weighted Blending achieved the lowest numerical error among the evaluated models on the testing subset, with an RMSE of 5.5403 dB, MAE of 3.7150 dB, and R² of 0.7967. Compared with average blending, the proposed model reduced RMSE by approximately 2.08%. The optimized weights indicated that XGBoost and MLP provided the largest contribution within the ensemble, while lower-weight learners still contributed complementary prediction characteristics. However, the improvement over XGBoost was relatively small and not statistically significant, indicating that the proposed framework should be interpreted as an adaptive ensemble strategy rather than a definitive replacement for the strongest individual learner.