A Dynamic Road Damage Prediction Using XGBoost and ArcGIS: Enhancing Infrastructure Management

Haoxiang Liu

Co-Presenters: Individual Presentation

College: College of Business and Public Management

Major: Finance

Faculty Research Mentor: Dan Liu

Abstract:

Flood disasters significantly disrupt road infrastructure, posing challenges to transportation networks and disaster management systems. Traditional approaches to assessing flood-induced road damage often rely on static methods or heuristic models, which fail to adapt to the dynamic and uncertain nature of flood events. These methods are also limited in their ability to integrate real-time data and provide accurate, actionable insights for decision-making. To address these limitations, this study introduces a Dynamic Road Damage Prediction Framework (DRDPF) that integrates XGBoost and ArcGIS for flood disaster management, focusing on dynamic risk assessment and geospatial visualization.The framework leverages XGBoost to predict road damage probabilities by combining static road and geographical features with dynamic weather and water level data derived from sliding time windows. Real-time IoT sensor data, including rainfall and water levels, is incorporated to ensure the model remains responsive to evolving flood conditions. ArcGIS is employed for spatial analysis and visualization, enabling the creation of dynamic heatmaps to identify high-risk road segments and prioritize recovery efforts effectively. This integration provides a data-driven and context-sensitive approach to flood risk management, ensuring timely and informed decision-making.To evaluate the effectiveness of DRDPF, the study focuses on a flood-prone region in New Jersey, utilizing comprehensive data from historical and real-time sources. The experimentation framework includes spatial and temporal cross-validation to ensure model robustness and generalizability. The results highlight the framework’s ability to accurately predict road damage probabilities and dynamically visualize risk distributions, demonstrating its potential to outperform traditional static methods in terms of adaptability and precision. By integrating XGBoost for predictive modeling and ArcGIS for geospatial analysis, the DRDPF offers a streamlined and scalable solution for flood disaster management, with significant implications for infrastructure resilience and emergency response planning.