Resource-Constrained Comparative Analysis of Road Asset Management Decision Frameworks

Abstract

Rural roads are essential for economic development in lower-income countries, yet many upgrading projects receive funding despite generating negative economic returns. This study empirically evaluates four decision-support approaches: condition-based heuristic ranking by International Roughness Index (IRI), multi-criteria analysis (MCA), PCA-based clustering, and integer linear programming (ILP) optimization. The evaluation uses a dataset of 166 rural roads in Sri Lanka and measures performance by accuracy (percentage of optimal net present value achieved) and overcapitalization risk (selection of negative-NPV roads) across six budget levels from 25% to 100% of the viable network investment. The condition-only heuristic achieves only 18–54% accuracy and consistently selects negative-NPV roads. MCA reaches 69–92% accuracy; with economic weighting (benefit-cost ratio at 50%) it attains 93%. PCA-clustering incorporating full economic data achieves 87–99% accuracy at constrained budgets, with zero overcapitalization up to 40% budget. Full ILP optimization never selects a negative-NPV road, but its marginal gain over clustering is small (1.75% per unit resource). Simpler frameworks serve equal or larger populations per investment than the NPV-optimum, revealing a trade-off between economic efficiency and population coverage. A sensitivity analysis shows that excluding all economic data from clustering reduces accuracy by 20–35 percentage points, while retaining only the benefit-cost ratio recovers most of this loss. A pragmatic decision rule is derived: in networks where IRI correlates weakly with economic return, condition-only heuristics should be avoided; full-profile clustering or economically weighted MCA deliver near-optimal performance at low cost; full optimization is justified only when economic data are already available.