Explainable Discrete-Time Survival Learning for Bridge Deck Deterioration and Risk-Based Preservation Prioritization

Shuxin Zhang; Lei Qiu

doi:10.6919/

Authors

Shuxin Zhang University of California, Berkeley, California, The United States
Lei Qiu Ningbo University of Technology, Ningbo, Zhejiang, China

DOI:

https://doi.org/10.6919/

Keywords:

Bridge management, deterioration modelling, survival analysis, discrete-time hazard, gradient boosting, SHAP, explainable machine learning, National Bridge Inventory, preservation prioritization.

Abstract

Highway agencies must decide which bridge decks to treat before their condition degrades, yet most data-driven deterioration models either impose restrictive parametric hazard shapes or recast the problem as snapshot classification of the current condition rating, which discards the censoring structure of periodic inspections. This paper presents an explainable survival machine-learning framework built directly on the sojourn-time structure of National Bridge Inventory (NBI) deck condition ratings. From a 23-year (1992–2014) nationwide panel of 150,144 concrete highway bridge decks we extract 48,901 right-censored sojourn spells in the "Good" state (condition rating 7), of which 16,182 end in an observed downgrade. We propose DT-GBH, a discrete-time gradient-boosted hazard model that expands every spell into person-period records and learns the annual conditional downgrade probability, so that interval-censored annual reporting, a fully flexible baseline hazard and non-proportional covariate effects are all handled natively. Against Kaplan–Meier, Cox proportional hazards, componentwise boosted Cox, random survival forest and XGBoost Cox/AFT baselines, DT-GBH gives the best discrimination (C-index 0.766, 95% CI [0.758, 0.776]; Uno's C 0.769; five-year AUC 0.784) and the lowest integrated Brier score (0.143). Because the model is a tree ensemble on the hazard logit, TreeSHAP provides exact additive attributions, and we introduce a time-resolved attribution view that exposes how the dominant drivers change with the years already spent in the state. Two methodological findings are reported: a uniform five-year retrospective window of deterioration-history features adds 0.033 C-index over inventory attributes alone, whereas calendar "panel-position" features inflate apparent accuracy by 0.013 and must be excluded; and transfer to unseen states degrades every model to C ≈ 0.66, where the machine-learning advantage over Cox disappears. Finally, predicted five-year downgrade probabilities are combined with traffic exposure and a normalized treatment-cost model into a Preservation Prioritization Index, which is scored retrospectively against transitions actually observed in held-out data: averaged over budget levels from 5% to 50% it recovers 82.0% of the benefit attainable by a clairvoyant ranking, against 51.6% for predicted risk alone and only 20.6% for the worst-first heuristic, which is barely better than random selection.

Downloads

Download data is not yet available.

References

[1] American Society of Civil Engineers. (2025). 2025 report card for America's infrastructure. Reston, VA: ASCE.

[2] Federal Highway Administration. (1995). Recording and coding guide for the structure inventory and appraisal of the nation's bridges (Rep. FHWA-PD-96-001). Washington, DC: FHWA.

[3] Federal Highway Administration. (2022). Specifications for the national bridge inventory. Washington, DC: FHWA.

[4] Morcous, G. (2006). Performance prediction of bridge deck systems using Markov chains. Journal of Performance of Constructed Facilities, 20(2), 146–155.

[5] Madanat, S., Mishalani, R., & Ibrahim, W. H. W. (1995). Estimation of infrastructure transition probabilities from condition rating data. Journal of Infrastructure Systems, 1(2), 120–125.

[6] Madanat, S., Karlaftis, M. G., & McCarthy, P. S. (1997). Probabilistic infrastructure deterioration models with panel data. Journal of Infrastructure Systems, 3(1), 4–9.

[7] Ng, S.-K., & Moses, F. (1998). Bridge deterioration modeling using semi-Markov theory. In Structural safety and reliability: Proc. ICOSSAR'97, Kyoto, Japan. Rotterdam: Balkema.

[8] Sobanjo, J. O. (2011). State transition probabilities in bridge deterioration based on Weibull sojourn times. Structure and Infrastructure Engineering, 7(10), 747–764.

[9] Thomas, O., & Sobanjo, J. (2016). Semi-Markov models for the deterioration of bridge elements. Journal of Infrastructure Systems, 22(3), 04016010.

[10] Thompson, P. D., & Johnson, M. B. (2005). Markovian bridge deterioration: Developing models from historical data. Structure and Infrastructure Engineering, 1(1), 85–91.

[11] Mauch, M., & Madanat, S. (2001). Semiparametric hazard rate models of reinforced concrete bridge deck deterioration. Journal of Infrastructure Systems, 7(2), 49–57.

[12] Mishalani, R., & Madanat, S. (2002). Computation of infrastructure transition probabilities using stochastic duration models. Journal of Infrastructure Systems, 8(4), 139–148.

[13] Agrawal, A. K., Kawaguchi, A., & Chen, Z. (2010). Deterioration rates of typical bridge elements in New York. Journal of Bridge Engineering, 15(4), 419–429.

[14] Sobanjo, J., Mtenga, P., & Rambo-Roddenberry, M. (2010). Reliability-based modeling of bridge deterioration hazards. Journal of Bridge Engineering, 15(6), 671–683.

[15] Manafpour, A., Guler, I., Radlińska, A., Rajabipour, F., & Warn, G. (2018). Stochastic analysis and time-based modeling of concrete bridge deck deterioration. Journal of Bridge Engineering, 23(9), 04018066.

[16] Fleischhacker, A., Ghonima, O., & Schumacher, T. (2020). Bayesian survival analysis for US concrete highway bridge decks. Journal of Infrastructure Systems, 26(1), 04020001.

[17] Ghonima, O., Anderson, J. C., Schumacher, T., & Unnikrishnan, A. (2020). Performance of US concrete highway bridge decks characterized by random parameters binary logistic regression. ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part A: Civil Engineering, 6(1). https://doi.org/10.1061/AJRUA6.0001031

[18] Bolukbasi, M., Mohammadi, J., & Arditi, D. (2004). Estimating the future condition of highway bridge components using national bridge inventory data. Practice Periodical on Structural Design and Construction, 9(1), 16–25.

[19] Bektas, B. A., Carriquiry, A., & Smadi, O. (2013). Using classification trees for predicting national bridge inventory condition ratings. Journal of Infrastructure Systems, 19(4), 425–433.

[20] Assaad, R., & El-adaway, I. H. (2020). Bridge infrastructure asset management system: Comparative computational machine learning approach for evaluating and predicting deck deterioration conditions. Journal of Infrastructure Systems, 26(3), 04020032.

[21] Li, Y., & Song, G. (2022). Ensemble-learning-based prediction of steel bridge deck defect condition. Applied Sciences, 12(11), 5442.

[22] Yang, C., Wang, X., & Nassif, H. (2024). Impact of environmental conditions on predicting condition rating of concrete bridge decks. Transportation Research Record. https://doi.org/10.1177/03611981241248647

[23] Chang, M., Maguire, M., & Sun, Y. (2017). Framework for mitigating human bias in the selection of explanatory variables for bridge deterioration modeling. Journal of Infrastructure Systems, 23(3), 04017002.

[24] Wang, F., Lee, C.-C., & Gharaibeh, N. G. (2022). Network-level bridge deterioration prediction models that consider the effect of maintenance and rehabilitation. Journal of Infrastructure Systems, 28, 05021009.

[25] Lu, M., & Guler, S. I. (2022). Comparison of random survival forest with accelerated failure time-Weibull model for bridge deck deterioration. Transportation Research Record, 2676(7), 296–311.

[26] Li, L., Lu, Y., & Peng, M. (2022). Deterioration model for reinforced concrete bridge girders based on survival analysis. Mathematics, 10(23), 4436.

[27] Huang, L., et al. (2022). A prognostic model for newly operated highway bridges based on censored data and survival analysis. Advances in Civil Engineering, 2022, 4667231.

[28] Cox, D. R. (1972). Regression models and life-tables. Journal of the Royal Statistical Society: Series B (Methodological), 34(2), 187–220.

[29] Kaplan, E. L., & Meier, P. (1958). Nonparametric estimation from incomplete observations. Journal of the American Statistical Association, 53(282), 457–481.

[30] Allison, P. D. (1982). Discrete-time methods for the analysis of event histories. Sociological Methodology, 13, 61–98.

[31] Tutz, G., & Schmid, M. (2016). Modeling discrete time-to-event data. Cham, Switzerland: Springer.

[32] Ishwaran, H., Kogalur, U. B., Blackstone, E. H., & Lauer, M. S. (2008). Random survival forests. The Annals of Applied Statistics, 2(3), 841–860.

[33] Berger, M., Welchowski, T., Schmitz-Valckenberg, S., & Schmid, M. (2019). A classification tree approach for the modeling of competing risks in discrete time. Advances in Data Analysis and Classification, 13(4), 965–990.

[34] Chen, T., & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 785–794).

[35] Ke, G., et al. (2017). LightGBM: A highly efficient gradient boosting decision tree. Advances in Neural Information Processing Systems, 30, 3146–3154.

[36] Katzman, J. L., Shaham, U., Cloninger, A., Bates, J., Jiang, T., & Kluger, Y. (2018). DeepSurv: Personalized treatment recommender system using a Cox proportional hazards deep neural network. BMC Medical Research Methodology, 18(1), 24.

[37] Kvamme, H., Borgan, Ø., & Scheel, I. (2019). Time-to-event prediction with neural networks and Cox regression. Journal of Machine Learning Research, 20(129), 1–30.

[38] Lundberg, S. M., & Lee, S.-I. (2017). A unified approach to interpreting model predictions. Advances in Neural Information Processing Systems, 30, 4765–4774.

[39] Lundberg, S. M., et al. (2020). From local explanations to global understanding with explainable AI for trees. Nature Machine Intelligence, 2(1), 56–67.

[40] Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). “Why should I trust you?” Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (pp. 1135–1144).

[41] Rudin, C. (2919). Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence, 1(5), 206–215.

[42] Harrell, F. E., Califf, R. M., Pryor, D. B., Lee, K. L., & Rosati, R. A. (1982). Evaluating the yield of medical tests. JAMA, 247(18), 2543–2546.

[43] Uno, H., Cai, T., Pencina, M. J., D'Agostino, R. B., & Wei, L. J. (2011). On the C-statistics for evaluating overall adequacy of risk prediction procedures with censored survival data. Statistics in Medicine, 30(10), 1105–1117.

[44] Graf, E., Schmoor, C., Sauerbrei, W., & Schumacher, M. (1999). Assessment and comparison of prognostic classification schemes for survival data. Statistics in Medicine, 18(17–18), 2529–2545.

[45] Pölsterl, S. (2020). scikit-survival: A library for time-to-event analysis built on top of scikit-learn. Journal of Machine Learning Research, 21(212), 1–6.

[46] Davidson-Pilon, C. (2019). lifelines: Survival analysis in Python. Journal of Open Source Software, 4(40), 1317.

[47] Pedregosa, F., et al. (2011). Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12, 2825–2830.

[48] Lu, M., Hydock, J., Radlińska, A., & Guler, S. I. (2022). Reliability analysis of a bridge deck utilizing generalized gamma distribution. Journal of Bridge Engineering, 27(4), 04022006.

[49] Frangopol, D. M., Kong, J. S., & Gharaibeh, E. S. (2001). Reliability-based life-cycle management of highway bridges. Journal of Computing in Civil Engineering, 15(1), 27–34.

[50] Almarahlleh, N., Liu, H., Abudayyeh, O., & Almamlook, R. (2024). Predicting concrete bridge deck deterioration: A hyperparameter optimization approach. Journal of Performance of Constructed Facilities, 38(3). https://doi.org/10.1061/JPCFEV.CFENG-4714

[51] Krzyziński, M., Spytek, M., Baniecki, H., & Biecek, P. (2023). SurvSHAP(t): Time-dependent explanations of machine learning survival models. Knowledge-Based Systems, 262, 110234.

[52] Kovalev, M. S., Utkin, L. V., & Kasimov, E. M. (2020). SurvLIME: A method for explaining machine learning survival models. Knowledge-Based Systems, 203, 106164.