Vaishnavi Ghogare, Ashwini Bhavar, Rohan Shelke

DOI: https://doi.org/10.5281/zenodo.22647917

Potholes are a major form of road deterioration that can affect road safety, vehicle performance, traffic movement, and transportation infrastructure. Traditional road inspection mainly depends on manual surveys and complaint-based reporting, which can be time-consuming, costly, and difficult to apply across large road networks. This paper proposes an integrated Artificial Intelligence (AI) and Geographic Information System (GIS)-based framework for automated pothole detection, severity assessment, repair-priority ranking, and approximate cost prediction. The framework uses YOLO-based computer vision to detect potholes from road images and extract visual characteristics including bounding-box width, height, area, aspect ratio, and relative image coverage. These features are intended for machine-learning models such as Random Forest and XGBoost to classify potholes into Low, Medium, and High severity categories. GIS-based analysis then incorporates contextual information such as road class, traffic conditions, proximity to critical services, and pothole distribution. Multi-Criteria Decision Analysis (MCDA) combines severity and spatial risk factors to generate repair-priority rankings. Regression models are proposed for approximate repair-cost prediction using defect characteristics, repair category, and road type. The current cleaned dataset contains 1,740 pothole records with image and bounding-box information. Because actual repair-cost labels are not present in the current dataset, cost prediction is defined as a subsequent stage requiring supplementary maintenance-cost records. The main contribution is a unified decision-support framework connecting visual detection, severity assessment, geographic context, repair prioritization, and cost planning.

Pothole Detection, Artificial Intelligence, Computer Vision, YOLO, Machine Learning, GIS, Random Forest, XGBoost, MCDA, Road Maintenance.

This paper presents an integrated AI and GIS-based framework for pothole severity assessment, repair-priority ranking, and approximate cost prediction. The proposed pipeline combines YOLO-based visual detection, bounding-box feature extraction, Random Forest and XGBoost severity classification, GIS contextual analysis, MCDA-based prioritization, and regression-based cost estimation. The current 1,740-record pothole dataset provides a foundation for the visual-feature and machine-learning stages. However, severity labels, GIS attributes, and actual repair-cost records must be incorporated before those components can be experimentally validated. The proposed framework therefore provides a practical and research-oriented foundation for intelligent road-maintenance decision support while clearly distinguishing the current evidence from future experimental outcomes.

J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You Only Look Once: Unified, Real-Time Object Detection,” Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), pp. 779–788, 2016.
[2] L. Breiman, “Random Forests,” Machine Learning, vol. 45, pp. 5–32, 2001.
[3] T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” Proc. 22nd ACM SIGKDD Int. Conf. Knowledge Discovery and Data Mining, pp. 785–794, 2016.
[4] D. Arya et al., “RDD2022: A multi-national image dataset for automatic Road Damage Detection,” arXiv:2209.08538, 2022.
[5] D. Arya et al., “RDD2020: An annotated image dataset for automatic road damage detection using deep learning,” Data in Brief, vol. 36, 107133, 2021.
[6] IEEE Author Center, “Structure Your Paper,” IEEE, 2026.
[7] IEEE Author Center, “Structure Your Article,” IEEE, 2026.

Vaishnavi Ghogare, Ashwini Bhavar, Rohan Shelke (2026). An AI and GIS-Based Framework for Pothole Severity Assessment, Repair Priority Ranking, and Cost Prediction. International Journal of Computer Techniques, 13(5), 334–337. ISSN: 2394-2231. DOI: https://doi.org/10.5281/zenodo.22647917