The development of traffic optimization projects in Ciudad Juárez faces the challenge of insufficient analytical metrics for accurate problem assessment. This study introduces a machine learning framework that automatically collects and generates synthetic traffic data from real travel times extracted from Google Maps through an automated Python routine using Selenium. The proposed Random Forest Regressor model was trained with temporal and spatial features such as coordinates, date, hour, and workday classification, enabling the prediction of total travel time with a mean absolute error below one minute and an overall accuracy of 91%. Empirical validation confirmed that the predicted travel times closely follow real observations, with deviations lower than 40 s on average. The results demonstrate that even with a limited set of variables, tree-based ensemble models can effectively learn traffic dynamics and generate reliable synthetic data for different urban conditions. The main innovation of this work lies in the low-cost, automated data acquisition method and its integration with predictive modeling to produce synthetic datasets that can support the design, calibration, and validation of future intelligent traffic management systems in Ciudad Juárez and other cities with limited mobility data. Additionally, the proposed approach establishes a methodological foundation for expanding the scope of traffic forecasting using artificial intelligence, combining accessibility, precision, and scalability within real-world urban environments.