Machine Learning-Enhanced Risk-Based Inspection Using XGBoost/LightGBM for Mechanical Integrity Management in Nigerian Oil and Gas Facilities

Abstract Most inspection tables we use in Nigeria were originally designed for temperate regions where a much slower decay rate is recorded compared to Nigeria where equipment tends to degrade five to eight times faster. This creates a dangerous mismatch, where we end up spending our limited inspection budgets on low-risk equipment while critical assets fail unexpectedly. Addressing this gap requires inspection tools calibrated to Nigerian operating realities rather than borrowed from foreign baselines. This paper offers an alternative. We built a Risk-Based Mechanical Integrity framework that replaces generic inspection tables with machine learning models (XGBoost and LightGBM) trained specifically on Nigerian field data. By feeding the models local corrosion drivers, environmental factors and actual maintenance history, we predict failure probabilities calibrated to Nigerian degradation conditions. The predictions are integrated with standard API 581 consequence assessment to set inspection priorities based on real-world risk. The framework was validated against 20,000 equipment records from Niger Delta facilities between 2020 and 2025. Validation yields an AUC-ROC of 0.858. Economic analysis of the validation dataset demonstrates inspection cost reductions of 30-40% and reduction in unplanned shutdowns by 60-70%. This framework provides Nigerian operators a data-driven tool for optimizing safety-critical inspection decisions within existing budget constraints.