Traffic noise is a spatially propagating form of environmental pollution that significantly affects public health, comfort, and environmental quality. Long-term exposure to traffic noise has been linked to cardiovascular diseases, sleep disturbance, and reduced mental well-being, as documented in epidemiological studies [19]. Traditional simulation tools for predicting traffic noise often rely on complex desktop-based software, demanding high computing power and specialized technical knowledge, which limits their accessibility and widespread use. In contrast, web-based applications are inherently lightweight and accessible across platforms. To bridge this gap, this study presents a methodology that combines geospatial machine learning and web-based GIS technologies to enable interactive scenario analysis of urban traffic noise exposure.The system employs a machine learning model, specifically LightGBM, trained on spatiotemporal traffic data and spatial configurations of urban environment features (roads and buildings) to estimate the environmental noise level at each receiver point in the city. By combining dynamic traffic variables with static spatial descriptors, the model achieves high predictive accuracy (RMSE = 0.9336, MAE = 0.5767) while maintaining computational efficiency suitable for web deployment. To enhance model transparency, SHAP analysis was conducted, revealing that vehicle–receiver distance and speed were the most influential predictors, consistent with physical expectations of noise propagation. Error analysis further indicated that higher prediction deviations occurred in areas with sparse traffic and low electric vehicle ratios. The application enables users to simulate different traffic scenarios by adjusting the electric vehicle ratio and instantly observe changes in noise levels and population-weighted exposure. Built with Vue3, Mapbox, and a Node.js–PostgreSQL backend, the platform supports smooth interaction, dynamic heatmap visualization, and multi-user access without installation. A comparative analysis of frontend and backend model deployment confirmed that backend deployment offers better scalability and stability for real-time multi-user predictions. Additionally, a user survey was conducted to evaluate the system’s usability and overall user experience.To assess how urban context influences the effectiveness of traffic noise mitigation, scenario-based experiments were conducted in two contrasting districts of Stockholm. The central research question was whether electrification yields different noise reduction outcomes depending on traffic speed and built environment characteristics. By using the trained ML model, incremental increases are simulated in electric vehicle (EV) ratios and changes are evaluated in predicted LAeq levels and population-weighted exposure. Results indicate that EV adoption leads to significantly greater noise reduction in low-speed, high-density areas compared to high-speed zones. These findings suggest that the benefits of electrification are spatially heterogeneous and that targeted, context-aware strategies are necessary for effective urban noise mitigation.