Investigating the impact of land use on urban traffic congestion via enhanced genetic algorithm optimized stacking ensemble learning: a case study in Xi’an

With the rapid pace of urbanization and the growing number of privately owned vehicles, traffic congestion has become a major obstacle to sustainable urban development. While most existing studies focus on short-term traffic state prediction, the long-term effects of land use characteristics on congestion remain underexplored. To address this gap, this study proposes an ensemble learning approach that combines an improved genetic algorithm with a Stacking framework to examine the complex relationship between land use patterns and traffic congestion. A neighborhood search mechanism is incorporated into the genetic algorithm to enhance local optimization and improve hyperparameter tuning, which is applied to base models such as eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Category Boosting (CatBoost). A Stacking ensemble model is then developed, using XGBoost as the meta-learner to integrate predictions from the base models, leading to improved classification performance. Experimental results show that the proposed method outperforms traditional models, particularly in handling nonlinear and imbalanced traffic data. SHapley Additive exPlanations (SHAP)-based interpretability analysis further reveals the significant role of key land use factors—especially the degree of residential-commercial land use mixing—in contributing to congestion, offering interpretable insights for urban planning. Overall, the results confirm the effectiveness of the improved genetic algorithm and demonstrate the potential of ensemble learning strategies in traffic prediction, providing a new methodological framework and theoretical basis for smart urban traffic management.