Conditional Axle Group Load Spectra from Short-Term WIM Data Using XGBoost: A Nairobi Case Study

Heavy and overloaded freight traffic strongly affects pavement performance, yet short-term weigh-in-motion (WIM) measurements are not easily converted into design-oriented traffic inputs. Using the Nairobi Southern Bypass in Kenya as a case study, this study develops axle load spectrum (ALS) and equivalent single axle load (ESAL) indicators from more than 1.5 million axle group records collected between June and December 2025 and proposes an XGBoost-based conditional axle load spectrum (CA-ALS) framework. The data revealed strongly right-skewed load distributions, with a limited number of heavily loaded axle groups dominating pavement damage. Compared with the static ALS by axle group type baseline, the CA-ALS reduced log loss from 2.7563 to 2.6709 in conditional spectrum prediction. In the December 2025 tandem axle benchmark, the CA-ALS increased the ESAL-based verification input by 6.0% at b = 4 and 11.1% at b = 5 relative to the stronger static reference. A legal-load-capped counterfactual analysis further showed that, for all heavy vehicles, observed overloading increased ESAL by 161.0% at b = 4 and 239.4% at b = 5. These results indicate that the CA-ALS provides condition-sensitive traffic inputs for design traffic verification, scenario-based pavement checks, and overload-sensitive evaluation based on short-term WIM observations.