The accelerated growth of vehicle fleets in Latin American cities, coupled with high altitudes and heavy traffic congestion, has substantially increased the environmental impact of carbon dioxide ( \(CO_2\) ) emissions. This work presents a practical methodology to predict \(CO_2\) emissions in urban areas, avoiding the need for computationally expensive traffic simulations. To achieve this, a dataset was generated by performing extensive microscopic traffic simulations with SUMO, using OpenStreetMap data to extract the urban road network and configuring vehicle flows based on official registration statistics from Ecuador. Multiple scenarios with varying vehicle densities and fleet compositions were simulated to build a diverse dataset. Three machine learning models, Linear Regression, Random Forest, and Neural Networks, were trained on this data set to predict \(CO_2\) emissions as a function of input traffic parameters. The Random Forest model outperformed the others, achieving \(R^2 = 0.9875\) and MAPE = 3.61%. This trained model was then deployed in a web application using Streamlit, allowing users to estimate emissions in real time by inputting simple traffic parameters, thereby eliminating the need for running new extensive SUMO simulations for each scenario. This framework offers an efficient decision support tool for urban planning and environmental assessment in high-altitude, traffic-congested cities like Quito.